ctl.c revision 268545
1/*-
2 * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3 * Copyright (c) 2012 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by Edward Tomasz Napierala
7 * under sponsorship from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions, and the following disclaimer,
14 *    without modification.
15 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
16 *    substantially similar to the "NO WARRANTY" disclaimer below
17 *    ("Disclaimer") and any redistribution must be conditioned upon
18 *    including a substantially similar Disclaimer requirement for further
19 *    binary redistribution.
20 *
21 * NO WARRANTY
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
30 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
31 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGES.
33 *
34 * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.c#8 $
35 */
36/*
37 * CAM Target Layer, a SCSI device emulation subsystem.
38 *
39 * Author: Ken Merry <ken@FreeBSD.org>
40 */
41
42#define _CTL_C
43
44#include <sys/cdefs.h>
45__FBSDID("$FreeBSD: stable/10/sys/cam/ctl/ctl.c 268545 2014-07-12 01:59:07Z mav $");
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/kernel.h>
50#include <sys/types.h>
51#include <sys/kthread.h>
52#include <sys/bio.h>
53#include <sys/fcntl.h>
54#include <sys/lock.h>
55#include <sys/module.h>
56#include <sys/mutex.h>
57#include <sys/condvar.h>
58#include <sys/malloc.h>
59#include <sys/conf.h>
60#include <sys/ioccom.h>
61#include <sys/queue.h>
62#include <sys/sbuf.h>
63#include <sys/smp.h>
64#include <sys/endian.h>
65#include <sys/sysctl.h>
66
67#include <cam/cam.h>
68#include <cam/scsi/scsi_all.h>
69#include <cam/scsi/scsi_da.h>
70#include <cam/ctl/ctl_io.h>
71#include <cam/ctl/ctl.h>
72#include <cam/ctl/ctl_frontend.h>
73#include <cam/ctl/ctl_frontend_internal.h>
74#include <cam/ctl/ctl_util.h>
75#include <cam/ctl/ctl_backend.h>
76#include <cam/ctl/ctl_ioctl.h>
77#include <cam/ctl/ctl_ha.h>
78#include <cam/ctl/ctl_private.h>
79#include <cam/ctl/ctl_debug.h>
80#include <cam/ctl/ctl_scsi_all.h>
81#include <cam/ctl/ctl_error.h>
82
83struct ctl_softc *control_softc = NULL;
84
85/*
86 * The default is to run with CTL_DONE_THREAD turned on.  Completed
87 * transactions are queued for processing by the CTL work thread.  When
88 * CTL_DONE_THREAD is not defined, completed transactions are processed in
89 * the caller's context.
90 */
91#define CTL_DONE_THREAD
92
93/*
94 * Use the serial number and device ID provided by the backend, rather than
95 * making up our own.
96 */
97#define CTL_USE_BACKEND_SN
98
99/*
100 * Size and alignment macros needed for Copan-specific HA hardware.  These
101 * can go away when the HA code is re-written, and uses busdma for any
102 * hardware.
103 */
104#define	CTL_ALIGN_8B(target, source, type)				\
105	if (((uint32_t)source & 0x7) != 0)				\
106		target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\
107	else								\
108		target = (type)source;
109
110#define	CTL_SIZE_8B(target, size)					\
111	if ((size & 0x7) != 0)						\
112		target = size + (0x8 - (size & 0x7));			\
113	else								\
114		target = size;
115
116#define CTL_ALIGN_8B_MARGIN	16
117
118/*
119 * Template mode pages.
120 */
121
122/*
123 * Note that these are default values only.  The actual values will be
124 * filled in when the user does a mode sense.
125 */
126static struct copan_power_subpage power_page_default = {
127	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
128	/*subpage*/ PWR_SUBPAGE_CODE,
129	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
130			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
131	/*page_version*/ PWR_VERSION,
132	/* total_luns */ 26,
133	/* max_active_luns*/ PWR_DFLT_MAX_LUNS,
134	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
135		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
136		      0, 0, 0, 0, 0, 0}
137};
138
139static struct copan_power_subpage power_page_changeable = {
140	/*page_code*/ PWR_PAGE_CODE | SMPH_SPF,
141	/*subpage*/ PWR_SUBPAGE_CODE,
142	/*page_length*/ {(sizeof(struct copan_power_subpage) - 4) & 0xff00,
143			 (sizeof(struct copan_power_subpage) - 4) & 0x00ff},
144	/*page_version*/ 0,
145	/* total_luns */ 0,
146	/* max_active_luns*/ 0,
147	/*reserved*/ {0, 0, 0, 0, 0, 0, 0, 0, 0,
148		      0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
149		      0, 0, 0, 0, 0, 0}
150};
151
152static struct copan_aps_subpage aps_page_default = {
153	APS_PAGE_CODE | SMPH_SPF, //page_code
154	APS_SUBPAGE_CODE, //subpage
155	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
156	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
157	APS_VERSION, //page_version
158	0, //lock_active
159	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
160	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
161	0, 0, 0, 0, 0} //reserved
162};
163
164static struct copan_aps_subpage aps_page_changeable = {
165	APS_PAGE_CODE | SMPH_SPF, //page_code
166	APS_SUBPAGE_CODE, //subpage
167	{(sizeof(struct copan_aps_subpage) - 4) & 0xff00,
168	 (sizeof(struct copan_aps_subpage) - 4) & 0x00ff}, //page_length
169	0, //page_version
170	0, //lock_active
171	{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
172	0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
173	0, 0, 0, 0, 0} //reserved
174};
175
176static struct copan_debugconf_subpage debugconf_page_default = {
177	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
178	DBGCNF_SUBPAGE_CODE,		/* subpage */
179	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
180	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
181	DBGCNF_VERSION,			/* page_version */
182	{CTL_TIME_IO_DEFAULT_SECS>>8,
183	 CTL_TIME_IO_DEFAULT_SECS>>0},	/* ctl_time_io_secs */
184};
185
186static struct copan_debugconf_subpage debugconf_page_changeable = {
187	DBGCNF_PAGE_CODE | SMPH_SPF,	/* page_code */
188	DBGCNF_SUBPAGE_CODE,		/* subpage */
189	{(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
190	 (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
191	0,				/* page_version */
192	{0xff,0xff},			/* ctl_time_io_secs */
193};
194
195static struct scsi_format_page format_page_default = {
196	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
197	/*page_length*/sizeof(struct scsi_format_page) - 2,
198	/*tracks_per_zone*/ {0, 0},
199	/*alt_sectors_per_zone*/ {0, 0},
200	/*alt_tracks_per_zone*/ {0, 0},
201	/*alt_tracks_per_lun*/ {0, 0},
202	/*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
203			        CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
204	/*bytes_per_sector*/ {0, 0},
205	/*interleave*/ {0, 0},
206	/*track_skew*/ {0, 0},
207	/*cylinder_skew*/ {0, 0},
208	/*flags*/ SFP_HSEC,
209	/*reserved*/ {0, 0, 0}
210};
211
212static struct scsi_format_page format_page_changeable = {
213	/*page_code*/SMS_FORMAT_DEVICE_PAGE,
214	/*page_length*/sizeof(struct scsi_format_page) - 2,
215	/*tracks_per_zone*/ {0, 0},
216	/*alt_sectors_per_zone*/ {0, 0},
217	/*alt_tracks_per_zone*/ {0, 0},
218	/*alt_tracks_per_lun*/ {0, 0},
219	/*sectors_per_track*/ {0, 0},
220	/*bytes_per_sector*/ {0, 0},
221	/*interleave*/ {0, 0},
222	/*track_skew*/ {0, 0},
223	/*cylinder_skew*/ {0, 0},
224	/*flags*/ 0,
225	/*reserved*/ {0, 0, 0}
226};
227
228static struct scsi_rigid_disk_page rigid_disk_page_default = {
229	/*page_code*/SMS_RIGID_DISK_PAGE,
230	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
231	/*cylinders*/ {0, 0, 0},
232	/*heads*/ CTL_DEFAULT_HEADS,
233	/*start_write_precomp*/ {0, 0, 0},
234	/*start_reduced_current*/ {0, 0, 0},
235	/*step_rate*/ {0, 0},
236	/*landing_zone_cylinder*/ {0, 0, 0},
237	/*rpl*/ SRDP_RPL_DISABLED,
238	/*rotational_offset*/ 0,
239	/*reserved1*/ 0,
240	/*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
241			   CTL_DEFAULT_ROTATION_RATE & 0xff},
242	/*reserved2*/ {0, 0}
243};
244
245static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
246	/*page_code*/SMS_RIGID_DISK_PAGE,
247	/*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
248	/*cylinders*/ {0, 0, 0},
249	/*heads*/ 0,
250	/*start_write_precomp*/ {0, 0, 0},
251	/*start_reduced_current*/ {0, 0, 0},
252	/*step_rate*/ {0, 0},
253	/*landing_zone_cylinder*/ {0, 0, 0},
254	/*rpl*/ 0,
255	/*rotational_offset*/ 0,
256	/*reserved1*/ 0,
257	/*rotation_rate*/ {0, 0},
258	/*reserved2*/ {0, 0}
259};
260
261static struct scsi_caching_page caching_page_default = {
262	/*page_code*/SMS_CACHING_PAGE,
263	/*page_length*/sizeof(struct scsi_caching_page) - 2,
264	/*flags1*/ SCP_DISC | SCP_WCE,
265	/*ret_priority*/ 0,
266	/*disable_pf_transfer_len*/ {0xff, 0xff},
267	/*min_prefetch*/ {0, 0},
268	/*max_prefetch*/ {0xff, 0xff},
269	/*max_pf_ceiling*/ {0xff, 0xff},
270	/*flags2*/ 0,
271	/*cache_segments*/ 0,
272	/*cache_seg_size*/ {0, 0},
273	/*reserved*/ 0,
274	/*non_cache_seg_size*/ {0, 0, 0}
275};
276
277static struct scsi_caching_page caching_page_changeable = {
278	/*page_code*/SMS_CACHING_PAGE,
279	/*page_length*/sizeof(struct scsi_caching_page) - 2,
280	/*flags1*/ 0,
281	/*ret_priority*/ 0,
282	/*disable_pf_transfer_len*/ {0, 0},
283	/*min_prefetch*/ {0, 0},
284	/*max_prefetch*/ {0, 0},
285	/*max_pf_ceiling*/ {0, 0},
286	/*flags2*/ 0,
287	/*cache_segments*/ 0,
288	/*cache_seg_size*/ {0, 0},
289	/*reserved*/ 0,
290	/*non_cache_seg_size*/ {0, 0, 0}
291};
292
293static struct scsi_control_page control_page_default = {
294	/*page_code*/SMS_CONTROL_MODE_PAGE,
295	/*page_length*/sizeof(struct scsi_control_page) - 2,
296	/*rlec*/0,
297	/*queue_flags*/0,
298	/*eca_and_aen*/0,
299	/*reserved*/0,
300	/*aen_holdoff_period*/{0, 0}
301};
302
303static struct scsi_control_page control_page_changeable = {
304	/*page_code*/SMS_CONTROL_MODE_PAGE,
305	/*page_length*/sizeof(struct scsi_control_page) - 2,
306	/*rlec*/SCP_DSENSE,
307	/*queue_flags*/0,
308	/*eca_and_aen*/0,
309	/*reserved*/0,
310	/*aen_holdoff_period*/{0, 0}
311};
312
313
314/*
315 * XXX KDM move these into the softc.
316 */
317static int rcv_sync_msg;
318static int persis_offset;
319static uint8_t ctl_pause_rtr;
320static int     ctl_is_single = 1;
321static int     index_to_aps_page;
322
323SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
324static int worker_threads = 1;
325TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
326SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
327    &worker_threads, 1, "Number of worker threads");
328static int verbose = 0;
329TUNABLE_INT("kern.cam.ctl.verbose", &verbose);
330SYSCTL_INT(_kern_cam_ctl, OID_AUTO, verbose, CTLFLAG_RWTUN,
331    &verbose, 0, "Show SCSI errors returned to initiator");
332
333/*
334 * Serial number (0x80), device id (0x83), supported pages (0x00),
335 * Block limits (0xB0) and Logical Block Provisioning (0xB2)
336 */
337#define SCSI_EVPD_NUM_SUPPORTED_PAGES	5
338
339static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
340				  int param);
341static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
342static int ctl_init(void);
343void ctl_shutdown(void);
344static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
345static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
346static void ctl_ioctl_online(void *arg);
347static void ctl_ioctl_offline(void *arg);
348static int ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id);
349static int ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id);
350static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id);
351static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id);
352static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio);
353static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock);
354static int ctl_ioctl_submit_wait(union ctl_io *io);
355static void ctl_ioctl_datamove(union ctl_io *io);
356static void ctl_ioctl_done(union ctl_io *io);
357static void ctl_ioctl_hard_startstop_callback(void *arg,
358					      struct cfi_metatask *metatask);
359static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask);
360static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
361			      struct ctl_ooa *ooa_hdr,
362			      struct ctl_ooa_entry *kern_entries);
363static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
364		     struct thread *td);
365uint32_t ctl_get_resindex(struct ctl_nexus *nexus);
366uint32_t ctl_port_idx(int port_num);
367#ifdef unused
368static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port,
369				   uint32_t targ_target, uint32_t targ_lun,
370				   int can_wait);
371static void ctl_kfree_io(union ctl_io *io);
372#endif /* unused */
373static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
374			 struct ctl_be_lun *be_lun, struct ctl_id target_id);
375static int ctl_free_lun(struct ctl_lun *lun);
376static void ctl_create_lun(struct ctl_be_lun *be_lun);
377/**
378static void ctl_failover_change_pages(struct ctl_softc *softc,
379				      struct ctl_scsiio *ctsio, int master);
380**/
381
382static int ctl_do_mode_select(union ctl_io *io);
383static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
384			   uint64_t res_key, uint64_t sa_res_key,
385			   uint8_t type, uint32_t residx,
386			   struct ctl_scsiio *ctsio,
387			   struct scsi_per_res_out *cdb,
388			   struct scsi_per_res_out_parms* param);
389static void ctl_pro_preempt_other(struct ctl_lun *lun,
390				  union ctl_ha_msg *msg);
391static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
392static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
393static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
394static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
395static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
396					 int alloc_len);
397static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
398static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
399static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
400static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len);
401static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2);
402static ctl_action ctl_check_for_blockage(union ctl_io *pending_io,
403					 union ctl_io *ooa_io);
404static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
405				union ctl_io *starting_io);
406static int ctl_check_blocked(struct ctl_lun *lun);
407static int ctl_scsiio_lun_check(struct ctl_softc *ctl_softc,
408				struct ctl_lun *lun,
409				struct ctl_cmd_entry *entry,
410				struct ctl_scsiio *ctsio);
411//static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc);
412static void ctl_failover(void);
413static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
414			       struct ctl_scsiio *ctsio);
415static int ctl_scsiio(struct ctl_scsiio *ctsio);
416
417static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
418static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
419			    ctl_ua_type ua_type);
420static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io,
421			 ctl_ua_type ua_type);
422static int ctl_abort_task(union ctl_io *io);
423static void ctl_run_task_queue(struct ctl_softc *ctl_softc);
424#ifdef CTL_IO_DELAY
425static void ctl_datamove_timer_wakeup(void *arg);
426static void ctl_done_timer_wakeup(void *arg);
427#endif /* CTL_IO_DELAY */
428
429static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
430static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
431static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
432static void ctl_datamove_remote_write(union ctl_io *io);
433static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
434static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
435static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
436static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
437				    ctl_ha_dt_cb callback);
438static void ctl_datamove_remote_read(union ctl_io *io);
439static void ctl_datamove_remote(union ctl_io *io);
440static int ctl_process_done(union ctl_io *io, int have_lock);
441static void ctl_work_thread(void *arg);
442
443/*
444 * Load the serialization table.  This isn't very pretty, but is probably
445 * the easiest way to do it.
446 */
447#include "ctl_ser_table.c"
448
449/*
450 * We only need to define open, close and ioctl routines for this driver.
451 */
452static struct cdevsw ctl_cdevsw = {
453	.d_version =	D_VERSION,
454	.d_flags =	0,
455	.d_open =	ctl_open,
456	.d_close =	ctl_close,
457	.d_ioctl =	ctl_ioctl,
458	.d_name =	"ctl",
459};
460
461
462MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
463
464static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
465
466static moduledata_t ctl_moduledata = {
467	"ctl",
468	ctl_module_event_handler,
469	NULL
470};
471
472DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
473MODULE_VERSION(ctl, 1);
474
475static void
476ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
477			    union ctl_ha_msg *msg_info)
478{
479	struct ctl_scsiio *ctsio;
480
481	if (msg_info->hdr.original_sc == NULL) {
482		printf("%s: original_sc == NULL!\n", __func__);
483		/* XXX KDM now what? */
484		return;
485	}
486
487	ctsio = &msg_info->hdr.original_sc->scsiio;
488	ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
489	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
490	ctsio->io_hdr.status = msg_info->hdr.status;
491	ctsio->scsi_status = msg_info->scsi.scsi_status;
492	ctsio->sense_len = msg_info->scsi.sense_len;
493	ctsio->sense_residual = msg_info->scsi.sense_residual;
494	ctsio->residual = msg_info->scsi.residual;
495	memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
496	       sizeof(ctsio->sense_data));
497	memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
498	       &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
499	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
500	ctl_wakeup_thread();
501}
502
503static void
504ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
505				union ctl_ha_msg *msg_info)
506{
507	struct ctl_scsiio *ctsio;
508
509	if (msg_info->hdr.serializing_sc == NULL) {
510		printf("%s: serializing_sc == NULL!\n", __func__);
511		/* XXX KDM now what? */
512		return;
513	}
514
515	ctsio = &msg_info->hdr.serializing_sc->scsiio;
516#if 0
517	/*
518	 * Attempt to catch the situation where an I/O has
519	 * been freed, and we're using it again.
520	 */
521	if (ctsio->io_hdr.io_type == 0xff) {
522		union ctl_io *tmp_io;
523		tmp_io = (union ctl_io *)ctsio;
524		printf("%s: %p use after free!\n", __func__,
525		       ctsio);
526		printf("%s: type %d msg %d cdb %x iptl: "
527		       "%d:%d:%d:%d tag 0x%04x "
528		       "flag %#x status %x\n",
529			__func__,
530			tmp_io->io_hdr.io_type,
531			tmp_io->io_hdr.msg_type,
532			tmp_io->scsiio.cdb[0],
533			tmp_io->io_hdr.nexus.initid.id,
534			tmp_io->io_hdr.nexus.targ_port,
535			tmp_io->io_hdr.nexus.targ_target.id,
536			tmp_io->io_hdr.nexus.targ_lun,
537			(tmp_io->io_hdr.io_type ==
538			CTL_IO_TASK) ?
539			tmp_io->taskio.tag_num :
540			tmp_io->scsiio.tag_num,
541		        tmp_io->io_hdr.flags,
542			tmp_io->io_hdr.status);
543	}
544#endif
545	ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
546	STAILQ_INSERT_TAIL(&ctl_softc->isc_queue, &ctsio->io_hdr, links);
547	ctl_wakeup_thread();
548}
549
550/*
551 * ISC (Inter Shelf Communication) event handler.  Events from the HA
552 * subsystem come in here.
553 */
554static void
555ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
556{
557	struct ctl_softc *ctl_softc;
558	union ctl_io *io;
559	struct ctl_prio *presio;
560	ctl_ha_status isc_status;
561
562	ctl_softc = control_softc;
563	io = NULL;
564
565
566#if 0
567	printf("CTL: Isc Msg event %d\n", event);
568#endif
569	if (event == CTL_HA_EVT_MSG_RECV) {
570		union ctl_ha_msg msg_info;
571
572		isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
573					     sizeof(msg_info), /*wait*/ 0);
574#if 0
575		printf("CTL: msg_type %d\n", msg_info.msg_type);
576#endif
577		if (isc_status != 0) {
578			printf("Error receiving message, status = %d\n",
579			       isc_status);
580			return;
581		}
582		mtx_lock(&ctl_softc->ctl_lock);
583
584		switch (msg_info.hdr.msg_type) {
585		case CTL_MSG_SERIALIZE:
586#if 0
587			printf("Serialize\n");
588#endif
589			io = ctl_alloc_io((void *)ctl_softc->othersc_pool);
590			if (io == NULL) {
591				printf("ctl_isc_event_handler: can't allocate "
592				       "ctl_io!\n");
593				/* Bad Juju */
594				/* Need to set busy and send msg back */
595				mtx_unlock(&ctl_softc->ctl_lock);
596				msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
597				msg_info.hdr.status = CTL_SCSI_ERROR;
598				msg_info.scsi.scsi_status = SCSI_STATUS_BUSY;
599				msg_info.scsi.sense_len = 0;
600			        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
601				    sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){
602				}
603				goto bailout;
604			}
605			ctl_zero_io(io);
606			// populate ctsio from msg_info
607			io->io_hdr.io_type = CTL_IO_SCSI;
608			io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
609			io->io_hdr.original_sc = msg_info.hdr.original_sc;
610#if 0
611			printf("pOrig %x\n", (int)msg_info.original_sc);
612#endif
613			io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
614					    CTL_FLAG_IO_ACTIVE;
615			/*
616			 * If we're in serialization-only mode, we don't
617			 * want to go through full done processing.  Thus
618			 * the COPY flag.
619			 *
620			 * XXX KDM add another flag that is more specific.
621			 */
622			if (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)
623				io->io_hdr.flags |= CTL_FLAG_INT_COPY;
624			io->io_hdr.nexus = msg_info.hdr.nexus;
625#if 0
626			printf("targ %d, port %d, iid %d, lun %d\n",
627			       io->io_hdr.nexus.targ_target.id,
628			       io->io_hdr.nexus.targ_port,
629			       io->io_hdr.nexus.initid.id,
630			       io->io_hdr.nexus.targ_lun);
631#endif
632			io->scsiio.tag_num = msg_info.scsi.tag_num;
633			io->scsiio.tag_type = msg_info.scsi.tag_type;
634			memcpy(io->scsiio.cdb, msg_info.scsi.cdb,
635			       CTL_MAX_CDBLEN);
636			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
637				struct ctl_cmd_entry *entry;
638				uint8_t opcode;
639
640				opcode = io->scsiio.cdb[0];
641				entry = &ctl_cmd_table[opcode];
642				io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
643				io->io_hdr.flags |=
644					entry->flags & CTL_FLAG_DATA_MASK;
645			}
646			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
647					   &io->io_hdr, links);
648			ctl_wakeup_thread();
649			break;
650
651		/* Performed on the Originating SC, XFER mode only */
652		case CTL_MSG_DATAMOVE: {
653			struct ctl_sg_entry *sgl;
654			int i, j;
655
656			io = msg_info.hdr.original_sc;
657			if (io == NULL) {
658				printf("%s: original_sc == NULL!\n", __func__);
659				/* XXX KDM do something here */
660				break;
661			}
662			io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
663			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
664			/*
665			 * Keep track of this, we need to send it back over
666			 * when the datamove is complete.
667			 */
668			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
669
670			if (msg_info.dt.sg_sequence == 0) {
671				/*
672				 * XXX KDM we use the preallocated S/G list
673				 * here, but we'll need to change this to
674				 * dynamic allocation if we need larger S/G
675				 * lists.
676				 */
677				if (msg_info.dt.kern_sg_entries >
678				    sizeof(io->io_hdr.remote_sglist) /
679				    sizeof(io->io_hdr.remote_sglist[0])) {
680					printf("%s: number of S/G entries "
681					    "needed %u > allocated num %zd\n",
682					    __func__,
683					    msg_info.dt.kern_sg_entries,
684					    sizeof(io->io_hdr.remote_sglist)/
685					    sizeof(io->io_hdr.remote_sglist[0]));
686
687					/*
688					 * XXX KDM send a message back to
689					 * the other side to shut down the
690					 * DMA.  The error will come back
691					 * through via the normal channel.
692					 */
693					break;
694				}
695				sgl = io->io_hdr.remote_sglist;
696				memset(sgl, 0,
697				       sizeof(io->io_hdr.remote_sglist));
698
699				io->scsiio.kern_data_ptr = (uint8_t *)sgl;
700
701				io->scsiio.kern_sg_entries =
702					msg_info.dt.kern_sg_entries;
703				io->scsiio.rem_sg_entries =
704					msg_info.dt.kern_sg_entries;
705				io->scsiio.kern_data_len =
706					msg_info.dt.kern_data_len;
707				io->scsiio.kern_total_len =
708					msg_info.dt.kern_total_len;
709				io->scsiio.kern_data_resid =
710					msg_info.dt.kern_data_resid;
711				io->scsiio.kern_rel_offset =
712					msg_info.dt.kern_rel_offset;
713				/*
714				 * Clear out per-DMA flags.
715				 */
716				io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK;
717				/*
718				 * Add per-DMA flags that are set for this
719				 * particular DMA request.
720				 */
721				io->io_hdr.flags |= msg_info.dt.flags &
722						    CTL_FLAG_RDMA_MASK;
723			} else
724				sgl = (struct ctl_sg_entry *)
725					io->scsiio.kern_data_ptr;
726
727			for (i = msg_info.dt.sent_sg_entries, j = 0;
728			     i < (msg_info.dt.sent_sg_entries +
729			     msg_info.dt.cur_sg_entries); i++, j++) {
730				sgl[i].addr = msg_info.dt.sg_list[j].addr;
731				sgl[i].len = msg_info.dt.sg_list[j].len;
732
733#if 0
734				printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
735				       __func__,
736				       msg_info.dt.sg_list[j].addr,
737				       msg_info.dt.sg_list[j].len,
738				       sgl[i].addr, sgl[i].len, j, i);
739#endif
740			}
741#if 0
742			memcpy(&sgl[msg_info.dt.sent_sg_entries],
743			       msg_info.dt.sg_list,
744			       sizeof(*sgl) * msg_info.dt.cur_sg_entries);
745#endif
746
747			/*
748			 * If this is the last piece of the I/O, we've got
749			 * the full S/G list.  Queue processing in the thread.
750			 * Otherwise wait for the next piece.
751			 */
752			if (msg_info.dt.sg_last != 0) {
753				STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
754						   &io->io_hdr, links);
755				ctl_wakeup_thread();
756			}
757			break;
758		}
759		/* Performed on the Serializing (primary) SC, XFER mode only */
760		case CTL_MSG_DATAMOVE_DONE: {
761			if (msg_info.hdr.serializing_sc == NULL) {
762				printf("%s: serializing_sc == NULL!\n",
763				       __func__);
764				/* XXX KDM now what? */
765				break;
766			}
767			/*
768			 * We grab the sense information here in case
769			 * there was a failure, so we can return status
770			 * back to the initiator.
771			 */
772			io = msg_info.hdr.serializing_sc;
773			io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
774			io->io_hdr.status = msg_info.hdr.status;
775			io->scsiio.scsi_status = msg_info.scsi.scsi_status;
776			io->scsiio.sense_len = msg_info.scsi.sense_len;
777			io->scsiio.sense_residual =msg_info.scsi.sense_residual;
778			io->io_hdr.port_status = msg_info.scsi.fetd_status;
779			io->scsiio.residual = msg_info.scsi.residual;
780			memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data,
781			       sizeof(io->scsiio.sense_data));
782
783			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
784					   &io->io_hdr, links);
785			ctl_wakeup_thread();
786			break;
787		}
788
789		/* Preformed on Originating SC, SER_ONLY mode */
790		case CTL_MSG_R2R:
791			io = msg_info.hdr.original_sc;
792			if (io == NULL) {
793				printf("%s: Major Bummer\n", __func__);
794				mtx_unlock(&ctl_softc->ctl_lock);
795				return;
796			} else {
797#if 0
798				printf("pOrig %x\n",(int) ctsio);
799#endif
800			}
801			io->io_hdr.msg_type = CTL_MSG_R2R;
802			io->io_hdr.serializing_sc = msg_info.hdr.serializing_sc;
803			STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
804					   &io->io_hdr, links);
805			ctl_wakeup_thread();
806			break;
807
808		/*
809		 * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
810		 * mode.
811		 * Performed on the Originating (i.e. secondary) SC in XFER
812		 * mode
813		 */
814		case CTL_MSG_FINISH_IO:
815			if (ctl_softc->ha_mode == CTL_HA_MODE_XFER)
816				ctl_isc_handler_finish_xfer(ctl_softc,
817							    &msg_info);
818			else
819				ctl_isc_handler_finish_ser_only(ctl_softc,
820								&msg_info);
821			break;
822
823		/* Preformed on Originating SC */
824		case CTL_MSG_BAD_JUJU:
825			io = msg_info.hdr.original_sc;
826			if (io == NULL) {
827				printf("%s: Bad JUJU!, original_sc is NULL!\n",
828				       __func__);
829				break;
830			}
831			ctl_copy_sense_data(&msg_info, io);
832			/*
833			 * IO should have already been cleaned up on other
834			 * SC so clear this flag so we won't send a message
835			 * back to finish the IO there.
836			 */
837			io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
838			io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
839
840			/* io = msg_info.hdr.serializing_sc; */
841			io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
842		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
843					   &io->io_hdr, links);
844			ctl_wakeup_thread();
845			break;
846
847		/* Handle resets sent from the other side */
848		case CTL_MSG_MANAGE_TASKS: {
849			struct ctl_taskio *taskio;
850			taskio = (struct ctl_taskio *)ctl_alloc_io(
851				(void *)ctl_softc->othersc_pool);
852			if (taskio == NULL) {
853				printf("ctl_isc_event_handler: can't allocate "
854				       "ctl_io!\n");
855				/* Bad Juju */
856				/* should I just call the proper reset func
857				   here??? */
858				mtx_unlock(&ctl_softc->ctl_lock);
859				goto bailout;
860			}
861			ctl_zero_io((union ctl_io *)taskio);
862			taskio->io_hdr.io_type = CTL_IO_TASK;
863			taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
864			taskio->io_hdr.nexus = msg_info.hdr.nexus;
865			taskio->task_action = msg_info.task.task_action;
866			taskio->tag_num = msg_info.task.tag_num;
867			taskio->tag_type = msg_info.task.tag_type;
868#ifdef CTL_TIME_IO
869			taskio->io_hdr.start_time = time_uptime;
870			getbintime(&taskio->io_hdr.start_bt);
871#if 0
872			cs_prof_gettime(&taskio->io_hdr.start_ticks);
873#endif
874#endif /* CTL_TIME_IO */
875		        STAILQ_INSERT_TAIL(&ctl_softc->task_queue,
876					   &taskio->io_hdr, links);
877			ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
878			ctl_wakeup_thread();
879			break;
880		}
881		/* Persistent Reserve action which needs attention */
882		case CTL_MSG_PERS_ACTION:
883			presio = (struct ctl_prio *)ctl_alloc_io(
884				(void *)ctl_softc->othersc_pool);
885			if (presio == NULL) {
886				printf("ctl_isc_event_handler: can't allocate "
887				       "ctl_io!\n");
888				/* Bad Juju */
889				/* Need to set busy and send msg back */
890				mtx_unlock(&ctl_softc->ctl_lock);
891				goto bailout;
892			}
893			ctl_zero_io((union ctl_io *)presio);
894			presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
895			presio->pr_msg = msg_info.pr;
896		        STAILQ_INSERT_TAIL(&ctl_softc->isc_queue,
897					   &presio->io_hdr, links);
898			ctl_wakeup_thread();
899			break;
900		case CTL_MSG_SYNC_FE:
901			rcv_sync_msg = 1;
902			break;
903		case CTL_MSG_APS_LOCK: {
904			// It's quicker to execute this then to
905			// queue it.
906			struct ctl_lun *lun;
907			struct ctl_page_index *page_index;
908			struct copan_aps_subpage *current_sp;
909			uint32_t targ_lun;
910
911			targ_lun = msg_info.hdr.nexus.targ_lun;
912			if (msg_info.hdr.nexus.lun_map_fn != NULL)
913				targ_lun = msg_info.hdr.nexus.lun_map_fn(msg_info.hdr.nexus.lun_map_arg, targ_lun);
914
915			lun = ctl_softc->ctl_luns[targ_lun];
916			page_index = &lun->mode_pages.index[index_to_aps_page];
917			current_sp = (struct copan_aps_subpage *)
918				     (page_index->page_data +
919				     (page_index->page_len * CTL_PAGE_CURRENT));
920
921			current_sp->lock_active = msg_info.aps.lock_flag;
922		        break;
923		}
924		default:
925		        printf("How did I get here?\n");
926		}
927		mtx_unlock(&ctl_softc->ctl_lock);
928	} else if (event == CTL_HA_EVT_MSG_SENT) {
929		if (param != CTL_HA_STATUS_SUCCESS) {
930			printf("Bad status from ctl_ha_msg_send status %d\n",
931			       param);
932		}
933		return;
934	} else if (event == CTL_HA_EVT_DISCONNECT) {
935		printf("CTL: Got a disconnect from Isc\n");
936		return;
937	} else {
938		printf("ctl_isc_event_handler: Unknown event %d\n", event);
939		return;
940	}
941
942bailout:
943	return;
944}
945
946static void
947ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
948{
949	struct scsi_sense_data *sense;
950
951	sense = &dest->scsiio.sense_data;
952	bcopy(&src->scsi.sense_data, sense, sizeof(*sense));
953	dest->scsiio.scsi_status = src->scsi.scsi_status;
954	dest->scsiio.sense_len = src->scsi.sense_len;
955	dest->io_hdr.status = src->hdr.status;
956}
957
958static int
959ctl_init(void)
960{
961	struct ctl_softc *softc;
962	struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool;
963	struct ctl_frontend *fe;
964        uint8_t sc_id =0;
965	int i, error, retval;
966	//int isc_retval;
967
968	retval = 0;
969	ctl_pause_rtr = 0;
970        rcv_sync_msg = 0;
971
972	control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
973			       M_WAITOK | M_ZERO);
974	softc = control_softc;
975
976	softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
977			      "cam/ctl");
978
979	softc->dev->si_drv1 = softc;
980
981	/*
982	 * By default, return a "bad LUN" peripheral qualifier for unknown
983	 * LUNs.  The user can override this default using the tunable or
984	 * sysctl.  See the comment in ctl_inquiry_std() for more details.
985	 */
986	softc->inquiry_pq_no_lun = 1;
987	TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun",
988			  &softc->inquiry_pq_no_lun);
989	sysctl_ctx_init(&softc->sysctl_ctx);
990	softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
991		SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
992		CTLFLAG_RD, 0, "CAM Target Layer");
993
994	if (softc->sysctl_tree == NULL) {
995		printf("%s: unable to allocate sysctl tree\n", __func__);
996		destroy_dev(softc->dev);
997		free(control_softc, M_DEVBUF);
998		control_softc = NULL;
999		return (ENOMEM);
1000	}
1001
1002	SYSCTL_ADD_INT(&softc->sysctl_ctx,
1003		       SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO,
1004		       "inquiry_pq_no_lun", CTLFLAG_RW,
1005		       &softc->inquiry_pq_no_lun, 0,
1006		       "Report no lun possible for invalid LUNs");
1007
1008	mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1009	mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF);
1010	softc->open_count = 0;
1011
1012	/*
1013	 * Default to actually sending a SYNCHRONIZE CACHE command down to
1014	 * the drive.
1015	 */
1016	softc->flags = CTL_FLAG_REAL_SYNC;
1017
1018	/*
1019	 * In Copan's HA scheme, the "master" and "slave" roles are
1020	 * figured out through the slot the controller is in.  Although it
1021	 * is an active/active system, someone has to be in charge.
1022 	 */
1023#ifdef NEEDTOPORT
1024        scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id);
1025#endif
1026
1027        if (sc_id == 0) {
1028		softc->flags |= CTL_FLAG_MASTER_SHELF;
1029		persis_offset = 0;
1030	} else
1031		persis_offset = CTL_MAX_INITIATORS;
1032
1033	/*
1034	 * XXX KDM need to figure out where we want to get our target ID
1035	 * and WWID.  Is it different on each port?
1036	 */
1037	softc->target.id = 0;
1038	softc->target.wwid[0] = 0x12345678;
1039	softc->target.wwid[1] = 0x87654321;
1040	STAILQ_INIT(&softc->lun_list);
1041	STAILQ_INIT(&softc->pending_lun_queue);
1042	STAILQ_INIT(&softc->task_queue);
1043	STAILQ_INIT(&softc->incoming_queue);
1044	STAILQ_INIT(&softc->rtr_queue);
1045	STAILQ_INIT(&softc->done_queue);
1046	STAILQ_INIT(&softc->isc_queue);
1047	STAILQ_INIT(&softc->fe_list);
1048	STAILQ_INIT(&softc->be_list);
1049	STAILQ_INIT(&softc->io_pools);
1050
1051	/*
1052	 * We don't bother calling these with ctl_lock held here, because,
1053	 * in theory, no one else can try to do anything while we're in our
1054	 * module init routine.
1055	 */
1056	if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL,
1057			    &internal_pool)!= 0){
1058		printf("ctl: can't allocate %d entry internal pool, "
1059		       "exiting\n", CTL_POOL_ENTRIES_INTERNAL);
1060		return (ENOMEM);
1061	}
1062
1063	if (ctl_pool_create(softc, CTL_POOL_EMERGENCY,
1064			    CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) {
1065		printf("ctl: can't allocate %d entry emergency pool, "
1066		       "exiting\n", CTL_POOL_ENTRIES_EMERGENCY);
1067		ctl_pool_free(internal_pool);
1068		return (ENOMEM);
1069	}
1070
1071	if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, CTL_POOL_ENTRIES_OTHER_SC,
1072	                    &other_pool) != 0)
1073	{
1074		printf("ctl: can't allocate %d entry other SC pool, "
1075		       "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1076		ctl_pool_free(internal_pool);
1077		ctl_pool_free(emergency_pool);
1078		return (ENOMEM);
1079	}
1080
1081	softc->internal_pool = internal_pool;
1082	softc->emergency_pool = emergency_pool;
1083	softc->othersc_pool = other_pool;
1084
1085	if (worker_threads > MAXCPU || worker_threads == 0) {
1086		printf("invalid kern.cam.ctl.worker_threads value; "
1087		    "setting to 1");
1088		worker_threads = 1;
1089	} else if (worker_threads < 0) {
1090		if (mp_ncpus > 2) {
1091			/*
1092			 * Using more than two worker threads actually hurts
1093			 * performance due to lock contention.
1094			 */
1095			worker_threads = 2;
1096		} else {
1097			worker_threads = 1;
1098		}
1099	}
1100
1101	for (i = 0; i < worker_threads; i++) {
1102		error = kproc_kthread_add(ctl_work_thread, softc,
1103		    &softc->work_thread, NULL, 0, 0, "ctl", "work%d", i);
1104		if (error != 0) {
1105			printf("error creating CTL work thread!\n");
1106			ctl_pool_free(internal_pool);
1107			ctl_pool_free(emergency_pool);
1108			ctl_pool_free(other_pool);
1109			return (error);
1110		}
1111	}
1112	if (bootverbose)
1113		printf("ctl: CAM Target Layer loaded\n");
1114
1115	/*
1116	 * Initialize the initiator and portname mappings
1117	 */
1118	memset(softc->wwpn_iid, 0, sizeof(softc->wwpn_iid));
1119
1120	/*
1121	 * Initialize the ioctl front end.
1122	 */
1123	fe = &softc->ioctl_info.fe;
1124	sprintf(softc->ioctl_info.port_name, "CTL ioctl");
1125	fe->port_type = CTL_PORT_IOCTL;
1126	fe->num_requested_ctl_io = 100;
1127	fe->port_name = softc->ioctl_info.port_name;
1128	fe->port_online = ctl_ioctl_online;
1129	fe->port_offline = ctl_ioctl_offline;
1130	fe->onoff_arg = &softc->ioctl_info;
1131	fe->targ_enable = ctl_ioctl_targ_enable;
1132	fe->targ_disable = ctl_ioctl_targ_disable;
1133	fe->lun_enable = ctl_ioctl_lun_enable;
1134	fe->lun_disable = ctl_ioctl_lun_disable;
1135	fe->targ_lun_arg = &softc->ioctl_info;
1136	fe->fe_datamove = ctl_ioctl_datamove;
1137	fe->fe_done = ctl_ioctl_done;
1138	fe->max_targets = 15;
1139	fe->max_target_id = 15;
1140
1141	if (ctl_frontend_register(&softc->ioctl_info.fe,
1142	                  (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) {
1143		printf("ctl: ioctl front end registration failed, will "
1144		       "continue anyway\n");
1145	}
1146
1147#ifdef CTL_IO_DELAY
1148	if (sizeof(struct callout) > CTL_TIMER_BYTES) {
1149		printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n",
1150		       sizeof(struct callout), CTL_TIMER_BYTES);
1151		return (EINVAL);
1152	}
1153#endif /* CTL_IO_DELAY */
1154
1155	return (0);
1156}
1157
1158void
1159ctl_shutdown(void)
1160{
1161	struct ctl_softc *softc;
1162	struct ctl_lun *lun, *next_lun;
1163	struct ctl_io_pool *pool;
1164
1165	softc = (struct ctl_softc *)control_softc;
1166
1167	if (ctl_frontend_deregister(&softc->ioctl_info.fe) != 0)
1168		printf("ctl: ioctl front end deregistration failed\n");
1169
1170	mtx_lock(&softc->ctl_lock);
1171
1172	/*
1173	 * Free up each LUN.
1174	 */
1175	for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1176		next_lun = STAILQ_NEXT(lun, links);
1177		ctl_free_lun(lun);
1178	}
1179
1180	mtx_unlock(&softc->ctl_lock);
1181
1182	/*
1183	 * This will rip the rug out from under any FETDs or anyone else
1184	 * that has a pool allocated.  Since we increment our module
1185	 * refcount any time someone outside the main CTL module allocates
1186	 * a pool, we shouldn't have any problems here.  The user won't be
1187	 * able to unload the CTL module until client modules have
1188	 * successfully unloaded.
1189	 */
1190	while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL)
1191		ctl_pool_free(pool);
1192
1193#if 0
1194	ctl_shutdown_thread(softc->work_thread);
1195#endif
1196
1197	mtx_destroy(&softc->pool_lock);
1198	mtx_destroy(&softc->ctl_lock);
1199
1200	destroy_dev(softc->dev);
1201
1202	sysctl_ctx_free(&softc->sysctl_ctx);
1203
1204	free(control_softc, M_DEVBUF);
1205	control_softc = NULL;
1206
1207	if (bootverbose)
1208		printf("ctl: CAM Target Layer unloaded\n");
1209}
1210
1211static int
1212ctl_module_event_handler(module_t mod, int what, void *arg)
1213{
1214
1215	switch (what) {
1216	case MOD_LOAD:
1217		return (ctl_init());
1218	case MOD_UNLOAD:
1219		return (EBUSY);
1220	default:
1221		return (EOPNOTSUPP);
1222	}
1223}
1224
1225/*
1226 * XXX KDM should we do some access checks here?  Bump a reference count to
1227 * prevent a CTL module from being unloaded while someone has it open?
1228 */
1229static int
1230ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1231{
1232	return (0);
1233}
1234
1235static int
1236ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1237{
1238	return (0);
1239}
1240
1241int
1242ctl_port_enable(ctl_port_type port_type)
1243{
1244	struct ctl_softc *softc;
1245	struct ctl_frontend *fe;
1246
1247	if (ctl_is_single == 0) {
1248		union ctl_ha_msg msg_info;
1249		int isc_retval;
1250
1251#if 0
1252		printf("%s: HA mode, synchronizing frontend enable\n",
1253		        __func__);
1254#endif
1255		msg_info.hdr.msg_type = CTL_MSG_SYNC_FE;
1256	        if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1257		        sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) {
1258			printf("Sync msg send error retval %d\n", isc_retval);
1259		}
1260		if (!rcv_sync_msg) {
1261			isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info,
1262			        sizeof(msg_info), 1);
1263		}
1264#if 0
1265        	printf("CTL:Frontend Enable\n");
1266	} else {
1267		printf("%s: single mode, skipping frontend synchronization\n",
1268		        __func__);
1269#endif
1270	}
1271
1272	softc = control_softc;
1273
1274	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1275		if (port_type & fe->port_type)
1276		{
1277#if 0
1278			printf("port %d\n", fe->targ_port);
1279#endif
1280			ctl_frontend_online(fe);
1281		}
1282	}
1283
1284	return (0);
1285}
1286
1287int
1288ctl_port_disable(ctl_port_type port_type)
1289{
1290	struct ctl_softc *softc;
1291	struct ctl_frontend *fe;
1292
1293	softc = control_softc;
1294
1295	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1296		if (port_type & fe->port_type)
1297			ctl_frontend_offline(fe);
1298	}
1299
1300	return (0);
1301}
1302
1303/*
1304 * Returns 0 for success, 1 for failure.
1305 * Currently the only failure mode is if there aren't enough entries
1306 * allocated.  So, in case of a failure, look at num_entries_dropped,
1307 * reallocate and try again.
1308 */
1309int
1310ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced,
1311	      int *num_entries_filled, int *num_entries_dropped,
1312	      ctl_port_type port_type, int no_virtual)
1313{
1314	struct ctl_softc *softc;
1315	struct ctl_frontend *fe;
1316	int entries_dropped, entries_filled;
1317	int retval;
1318	int i;
1319
1320	softc = control_softc;
1321
1322	retval = 0;
1323	entries_filled = 0;
1324	entries_dropped = 0;
1325
1326	i = 0;
1327	mtx_lock(&softc->ctl_lock);
1328	STAILQ_FOREACH(fe, &softc->fe_list, links) {
1329		struct ctl_port_entry *entry;
1330
1331		if ((fe->port_type & port_type) == 0)
1332			continue;
1333
1334		if ((no_virtual != 0)
1335		 && (fe->virtual_port != 0))
1336			continue;
1337
1338		if (entries_filled >= num_entries_alloced) {
1339			entries_dropped++;
1340			continue;
1341		}
1342		entry = &entries[i];
1343
1344		entry->port_type = fe->port_type;
1345		strlcpy(entry->port_name, fe->port_name,
1346			sizeof(entry->port_name));
1347		entry->physical_port = fe->physical_port;
1348		entry->virtual_port = fe->virtual_port;
1349		entry->wwnn = fe->wwnn;
1350		entry->wwpn = fe->wwpn;
1351
1352		i++;
1353		entries_filled++;
1354	}
1355
1356	mtx_unlock(&softc->ctl_lock);
1357
1358	if (entries_dropped > 0)
1359		retval = 1;
1360
1361	*num_entries_dropped = entries_dropped;
1362	*num_entries_filled = entries_filled;
1363
1364	return (retval);
1365}
1366
1367static void
1368ctl_ioctl_online(void *arg)
1369{
1370	struct ctl_ioctl_info *ioctl_info;
1371
1372	ioctl_info = (struct ctl_ioctl_info *)arg;
1373
1374	ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED;
1375}
1376
1377static void
1378ctl_ioctl_offline(void *arg)
1379{
1380	struct ctl_ioctl_info *ioctl_info;
1381
1382	ioctl_info = (struct ctl_ioctl_info *)arg;
1383
1384	ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED;
1385}
1386
1387/*
1388 * Remove an initiator by port number and initiator ID.
1389 * Returns 0 for success, 1 for failure.
1390 */
1391int
1392ctl_remove_initiator(int32_t targ_port, uint32_t iid)
1393{
1394	struct ctl_softc *softc;
1395
1396	softc = control_softc;
1397
1398	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1399
1400	if ((targ_port < 0)
1401	 || (targ_port > CTL_MAX_PORTS)) {
1402		printf("%s: invalid port number %d\n", __func__, targ_port);
1403		return (1);
1404	}
1405	if (iid > CTL_MAX_INIT_PER_PORT) {
1406		printf("%s: initiator ID %u > maximun %u!\n",
1407		       __func__, iid, CTL_MAX_INIT_PER_PORT);
1408		return (1);
1409	}
1410
1411	mtx_lock(&softc->ctl_lock);
1412
1413	softc->wwpn_iid[targ_port][iid].in_use = 0;
1414
1415	mtx_unlock(&softc->ctl_lock);
1416
1417	return (0);
1418}
1419
1420/*
1421 * Add an initiator to the initiator map.
1422 * Returns 0 for success, 1 for failure.
1423 */
1424int
1425ctl_add_initiator(uint64_t wwpn, int32_t targ_port, uint32_t iid)
1426{
1427	struct ctl_softc *softc;
1428	int retval;
1429
1430	softc = control_softc;
1431
1432	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1433
1434	retval = 0;
1435
1436	if ((targ_port < 0)
1437	 || (targ_port > CTL_MAX_PORTS)) {
1438		printf("%s: invalid port number %d\n", __func__, targ_port);
1439		return (1);
1440	}
1441	if (iid > CTL_MAX_INIT_PER_PORT) {
1442		printf("%s: WWPN %#jx initiator ID %u > maximun %u!\n",
1443		       __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1444		return (1);
1445	}
1446
1447	mtx_lock(&softc->ctl_lock);
1448
1449	if (softc->wwpn_iid[targ_port][iid].in_use != 0) {
1450		/*
1451		 * We don't treat this as an error.
1452		 */
1453		if (softc->wwpn_iid[targ_port][iid].wwpn == wwpn) {
1454			printf("%s: port %d iid %u WWPN %#jx arrived again?\n",
1455			       __func__, targ_port, iid, (uintmax_t)wwpn);
1456			goto bailout;
1457		}
1458
1459		/*
1460		 * This is an error, but what do we do about it?  The
1461		 * driver is telling us we have a new WWPN for this
1462		 * initiator ID, so we pretty much need to use it.
1463		 */
1464		printf("%s: port %d iid %u WWPN %#jx arrived, WWPN %#jx is "
1465		       "still at that address\n", __func__, targ_port, iid,
1466		       (uintmax_t)wwpn,
1467		       (uintmax_t)softc->wwpn_iid[targ_port][iid].wwpn);
1468
1469		/*
1470		 * XXX KDM clear have_ca and ua_pending on each LUN for
1471		 * this initiator.
1472		 */
1473	}
1474	softc->wwpn_iid[targ_port][iid].in_use = 1;
1475	softc->wwpn_iid[targ_port][iid].iid = iid;
1476	softc->wwpn_iid[targ_port][iid].wwpn = wwpn;
1477	softc->wwpn_iid[targ_port][iid].port = targ_port;
1478
1479bailout:
1480
1481	mtx_unlock(&softc->ctl_lock);
1482
1483	return (retval);
1484}
1485
1486/*
1487 * XXX KDM should we pretend to do something in the target/lun
1488 * enable/disable functions?
1489 */
1490static int
1491ctl_ioctl_targ_enable(void *arg, struct ctl_id targ_id)
1492{
1493	return (0);
1494}
1495
1496static int
1497ctl_ioctl_targ_disable(void *arg, struct ctl_id targ_id)
1498{
1499	return (0);
1500}
1501
1502static int
1503ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id)
1504{
1505	return (0);
1506}
1507
1508static int
1509ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id)
1510{
1511	return (0);
1512}
1513
1514/*
1515 * Data movement routine for the CTL ioctl frontend port.
1516 */
1517static int
1518ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio)
1519{
1520	struct ctl_sg_entry *ext_sglist, *kern_sglist;
1521	struct ctl_sg_entry ext_entry, kern_entry;
1522	int ext_sglen, ext_sg_entries, kern_sg_entries;
1523	int ext_sg_start, ext_offset;
1524	int len_to_copy, len_copied;
1525	int kern_watermark, ext_watermark;
1526	int ext_sglist_malloced;
1527	int i, j;
1528
1529	ext_sglist_malloced = 0;
1530	ext_sg_start = 0;
1531	ext_offset = 0;
1532
1533	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n"));
1534
1535	/*
1536	 * If this flag is set, fake the data transfer.
1537	 */
1538	if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) {
1539		ctsio->ext_data_filled = ctsio->ext_data_len;
1540		goto bailout;
1541	}
1542
1543	/*
1544	 * To simplify things here, if we have a single buffer, stick it in
1545	 * a S/G entry and just make it a single entry S/G list.
1546	 */
1547	if (ctsio->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) {
1548		int len_seen;
1549
1550		ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist);
1551
1552		ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL,
1553							   M_WAITOK);
1554		ext_sglist_malloced = 1;
1555		if (copyin(ctsio->ext_data_ptr, ext_sglist,
1556				   ext_sglen) != 0) {
1557			ctl_set_internal_failure(ctsio,
1558						 /*sks_valid*/ 0,
1559						 /*retry_count*/ 0);
1560			goto bailout;
1561		}
1562		ext_sg_entries = ctsio->ext_sg_entries;
1563		len_seen = 0;
1564		for (i = 0; i < ext_sg_entries; i++) {
1565			if ((len_seen + ext_sglist[i].len) >=
1566			     ctsio->ext_data_filled) {
1567				ext_sg_start = i;
1568				ext_offset = ctsio->ext_data_filled - len_seen;
1569				break;
1570			}
1571			len_seen += ext_sglist[i].len;
1572		}
1573	} else {
1574		ext_sglist = &ext_entry;
1575		ext_sglist->addr = ctsio->ext_data_ptr;
1576		ext_sglist->len = ctsio->ext_data_len;
1577		ext_sg_entries = 1;
1578		ext_sg_start = 0;
1579		ext_offset = ctsio->ext_data_filled;
1580	}
1581
1582	if (ctsio->kern_sg_entries > 0) {
1583		kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr;
1584		kern_sg_entries = ctsio->kern_sg_entries;
1585	} else {
1586		kern_sglist = &kern_entry;
1587		kern_sglist->addr = ctsio->kern_data_ptr;
1588		kern_sglist->len = ctsio->kern_data_len;
1589		kern_sg_entries = 1;
1590	}
1591
1592
1593	kern_watermark = 0;
1594	ext_watermark = ext_offset;
1595	len_copied = 0;
1596	for (i = ext_sg_start, j = 0;
1597	     i < ext_sg_entries && j < kern_sg_entries;) {
1598		uint8_t *ext_ptr, *kern_ptr;
1599
1600		len_to_copy = ctl_min(ext_sglist[i].len - ext_watermark,
1601				      kern_sglist[j].len - kern_watermark);
1602
1603		ext_ptr = (uint8_t *)ext_sglist[i].addr;
1604		ext_ptr = ext_ptr + ext_watermark;
1605		if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
1606			/*
1607			 * XXX KDM fix this!
1608			 */
1609			panic("need to implement bus address support");
1610#if 0
1611			kern_ptr = bus_to_virt(kern_sglist[j].addr);
1612#endif
1613		} else
1614			kern_ptr = (uint8_t *)kern_sglist[j].addr;
1615		kern_ptr = kern_ptr + kern_watermark;
1616
1617		kern_watermark += len_to_copy;
1618		ext_watermark += len_to_copy;
1619
1620		if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
1621		     CTL_FLAG_DATA_IN) {
1622			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1623					 "bytes to user\n", len_to_copy));
1624			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1625					 "to %p\n", kern_ptr, ext_ptr));
1626			if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) {
1627				ctl_set_internal_failure(ctsio,
1628							 /*sks_valid*/ 0,
1629							 /*retry_count*/ 0);
1630				goto bailout;
1631			}
1632		} else {
1633			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d "
1634					 "bytes from user\n", len_to_copy));
1635			CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p "
1636					 "to %p\n", ext_ptr, kern_ptr));
1637			if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){
1638				ctl_set_internal_failure(ctsio,
1639							 /*sks_valid*/ 0,
1640							 /*retry_count*/0);
1641				goto bailout;
1642			}
1643		}
1644
1645		len_copied += len_to_copy;
1646
1647		if (ext_sglist[i].len == ext_watermark) {
1648			i++;
1649			ext_watermark = 0;
1650		}
1651
1652		if (kern_sglist[j].len == kern_watermark) {
1653			j++;
1654			kern_watermark = 0;
1655		}
1656	}
1657
1658	ctsio->ext_data_filled += len_copied;
1659
1660	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, "
1661			 "kern_sg_entries: %d\n", ext_sg_entries,
1662			 kern_sg_entries));
1663	CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, "
1664			 "kern_data_len = %d\n", ctsio->ext_data_len,
1665			 ctsio->kern_data_len));
1666
1667
1668	/* XXX KDM set residual?? */
1669bailout:
1670
1671	if (ext_sglist_malloced != 0)
1672		free(ext_sglist, M_CTL);
1673
1674	return (CTL_RETVAL_COMPLETE);
1675}
1676
1677/*
1678 * Serialize a command that went down the "wrong" side, and so was sent to
1679 * this controller for execution.  The logic is a little different than the
1680 * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1681 * sent back to the other side, but in the success case, we execute the
1682 * command on this side (XFER mode) or tell the other side to execute it
1683 * (SER_ONLY mode).
1684 */
1685static int
1686ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio, int have_lock)
1687{
1688	struct ctl_softc *ctl_softc;
1689	union ctl_ha_msg msg_info;
1690	struct ctl_lun *lun;
1691	int retval = 0;
1692	uint32_t targ_lun;
1693
1694	ctl_softc = control_softc;
1695	if (have_lock == 0)
1696		mtx_lock(&ctl_softc->ctl_lock);
1697
1698	targ_lun = ctsio->io_hdr.nexus.targ_lun;
1699	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
1700		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
1701	lun = ctl_softc->ctl_luns[targ_lun];
1702	if (lun==NULL)
1703	{
1704		/*
1705		 * Why isn't LUN defined? The other side wouldn't
1706		 * send a cmd if the LUN is undefined.
1707		 */
1708		printf("%s: Bad JUJU!, LUN is NULL!\n", __func__);
1709
1710		/* "Logical unit not supported" */
1711		ctl_set_sense_data(&msg_info.scsi.sense_data,
1712				   lun,
1713				   /*sense_format*/SSD_TYPE_NONE,
1714				   /*current_error*/ 1,
1715				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1716				   /*asc*/ 0x25,
1717				   /*ascq*/ 0x00,
1718				   SSD_ELEM_NONE);
1719
1720		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1721		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1722		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1723		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1724		msg_info.hdr.serializing_sc = NULL;
1725		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1726	        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1727				sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1728		}
1729		if (have_lock == 0)
1730			mtx_unlock(&ctl_softc->ctl_lock);
1731		return(1);
1732
1733	}
1734
1735    	TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1736
1737	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1738		(union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1739		 ooa_links))) {
1740	case CTL_ACTION_BLOCK:
1741		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1742		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1743				  blocked_links);
1744		break;
1745	case CTL_ACTION_PASS:
1746	case CTL_ACTION_SKIP:
1747		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
1748			ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1749			STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
1750					   &ctsio->io_hdr, links);
1751		} else {
1752
1753			/* send msg back to other side */
1754			msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1755			msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1756			msg_info.hdr.msg_type = CTL_MSG_R2R;
1757#if 0
1758			printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc);
1759#endif
1760		        if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1761			    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1762			}
1763		}
1764		break;
1765	case CTL_ACTION_OVERLAP:
1766		/* OVERLAPPED COMMANDS ATTEMPTED */
1767		ctl_set_sense_data(&msg_info.scsi.sense_data,
1768				   lun,
1769				   /*sense_format*/SSD_TYPE_NONE,
1770				   /*current_error*/ 1,
1771				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1772				   /*asc*/ 0x4E,
1773				   /*ascq*/ 0x00,
1774				   SSD_ELEM_NONE);
1775
1776		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1777		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1778		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1779		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1780		msg_info.hdr.serializing_sc = NULL;
1781		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1782#if 0
1783		printf("BAD JUJU:Major Bummer Overlap\n");
1784#endif
1785		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1786		retval = 1;
1787		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1788		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1789		}
1790		break;
1791	case CTL_ACTION_OVERLAP_TAG:
1792		/* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */
1793		ctl_set_sense_data(&msg_info.scsi.sense_data,
1794				   lun,
1795				   /*sense_format*/SSD_TYPE_NONE,
1796				   /*current_error*/ 1,
1797				   /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
1798				   /*asc*/ 0x4D,
1799				   /*ascq*/ ctsio->tag_num & 0xff,
1800				   SSD_ELEM_NONE);
1801
1802		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1803		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1804		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1805		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1806		msg_info.hdr.serializing_sc = NULL;
1807		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1808#if 0
1809		printf("BAD JUJU:Major Bummer Overlap Tag\n");
1810#endif
1811		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1812		retval = 1;
1813		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1814		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1815		}
1816		break;
1817	case CTL_ACTION_ERROR:
1818	default:
1819		/* "Internal target failure" */
1820		ctl_set_sense_data(&msg_info.scsi.sense_data,
1821				   lun,
1822				   /*sense_format*/SSD_TYPE_NONE,
1823				   /*current_error*/ 1,
1824				   /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
1825				   /*asc*/ 0x44,
1826				   /*ascq*/ 0x00,
1827				   SSD_ELEM_NONE);
1828
1829		msg_info.scsi.sense_len = SSD_FULL_SIZE;
1830		msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
1831		msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
1832		msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1833		msg_info.hdr.serializing_sc = NULL;
1834		msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1835#if 0
1836		printf("BAD JUJU:Major Bummer HW Error\n");
1837#endif
1838		TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1839		retval = 1;
1840		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1841		    sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) {
1842		}
1843		break;
1844	}
1845	if (have_lock == 0)
1846		mtx_unlock(&ctl_softc->ctl_lock);
1847	return (retval);
1848}
1849
1850static int
1851ctl_ioctl_submit_wait(union ctl_io *io)
1852{
1853	struct ctl_fe_ioctl_params params;
1854	ctl_fe_ioctl_state last_state;
1855	int done, retval;
1856
1857	retval = 0;
1858
1859	bzero(&params, sizeof(params));
1860
1861	mtx_init(&params.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF);
1862	cv_init(&params.sem, "ctlioccv");
1863	params.state = CTL_IOCTL_INPROG;
1864	last_state = params.state;
1865
1866	io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = &params;
1867
1868	CTL_DEBUG_PRINT(("ctl_ioctl_submit_wait\n"));
1869
1870	/* This shouldn't happen */
1871	if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE)
1872		return (retval);
1873
1874	done = 0;
1875
1876	do {
1877		mtx_lock(&params.ioctl_mtx);
1878		/*
1879		 * Check the state here, and don't sleep if the state has
1880		 * already changed (i.e. wakeup has already occured, but we
1881		 * weren't waiting yet).
1882		 */
1883		if (params.state == last_state) {
1884			/* XXX KDM cv_wait_sig instead? */
1885			cv_wait(&params.sem, &params.ioctl_mtx);
1886		}
1887		last_state = params.state;
1888
1889		switch (params.state) {
1890		case CTL_IOCTL_INPROG:
1891			/* Why did we wake up? */
1892			/* XXX KDM error here? */
1893			mtx_unlock(&params.ioctl_mtx);
1894			break;
1895		case CTL_IOCTL_DATAMOVE:
1896			CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n"));
1897
1898			/*
1899			 * change last_state back to INPROG to avoid
1900			 * deadlock on subsequent data moves.
1901			 */
1902			params.state = last_state = CTL_IOCTL_INPROG;
1903
1904			mtx_unlock(&params.ioctl_mtx);
1905			ctl_ioctl_do_datamove(&io->scsiio);
1906			/*
1907			 * Note that in some cases, most notably writes,
1908			 * this will queue the I/O and call us back later.
1909			 * In other cases, generally reads, this routine
1910			 * will immediately call back and wake us up,
1911			 * probably using our own context.
1912			 */
1913			io->scsiio.be_move_done(io);
1914			break;
1915		case CTL_IOCTL_DONE:
1916			mtx_unlock(&params.ioctl_mtx);
1917			CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n"));
1918			done = 1;
1919			break;
1920		default:
1921			mtx_unlock(&params.ioctl_mtx);
1922			/* XXX KDM error here? */
1923			break;
1924		}
1925	} while (done == 0);
1926
1927	mtx_destroy(&params.ioctl_mtx);
1928	cv_destroy(&params.sem);
1929
1930	return (CTL_RETVAL_COMPLETE);
1931}
1932
1933static void
1934ctl_ioctl_datamove(union ctl_io *io)
1935{
1936	struct ctl_fe_ioctl_params *params;
1937
1938	params = (struct ctl_fe_ioctl_params *)
1939		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1940
1941	mtx_lock(&params->ioctl_mtx);
1942	params->state = CTL_IOCTL_DATAMOVE;
1943	cv_broadcast(&params->sem);
1944	mtx_unlock(&params->ioctl_mtx);
1945}
1946
1947static void
1948ctl_ioctl_done(union ctl_io *io)
1949{
1950	struct ctl_fe_ioctl_params *params;
1951
1952	params = (struct ctl_fe_ioctl_params *)
1953		io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr;
1954
1955	mtx_lock(&params->ioctl_mtx);
1956	params->state = CTL_IOCTL_DONE;
1957	cv_broadcast(&params->sem);
1958	mtx_unlock(&params->ioctl_mtx);
1959}
1960
1961static void
1962ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask)
1963{
1964	struct ctl_fe_ioctl_startstop_info *sd_info;
1965
1966	sd_info = (struct ctl_fe_ioctl_startstop_info *)arg;
1967
1968	sd_info->hs_info.status = metatask->status;
1969	sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns;
1970	sd_info->hs_info.luns_complete =
1971		metatask->taskinfo.startstop.luns_complete;
1972	sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed;
1973
1974	cv_broadcast(&sd_info->sem);
1975}
1976
1977static void
1978ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask)
1979{
1980	struct ctl_fe_ioctl_bbrread_info *fe_bbr_info;
1981
1982	fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg;
1983
1984	mtx_lock(fe_bbr_info->lock);
1985	fe_bbr_info->bbr_info->status = metatask->status;
1986	fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
1987	fe_bbr_info->wakeup_done = 1;
1988	mtx_unlock(fe_bbr_info->lock);
1989
1990	cv_broadcast(&fe_bbr_info->sem);
1991}
1992
1993/*
1994 * Returns 0 for success, errno for failure.
1995 */
1996static int
1997ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1998		   struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1999{
2000	union ctl_io *io;
2001	int retval;
2002
2003	retval = 0;
2004
2005	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
2006
2007	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2008	     (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2009	     ooa_links)) {
2010		struct ctl_ooa_entry *entry;
2011
2012		/*
2013		 * If we've got more than we can fit, just count the
2014		 * remaining entries.
2015		 */
2016		if (*cur_fill_num >= ooa_hdr->alloc_num)
2017			continue;
2018
2019		entry = &kern_entries[*cur_fill_num];
2020
2021		entry->tag_num = io->scsiio.tag_num;
2022		entry->lun_num = lun->lun;
2023#ifdef CTL_TIME_IO
2024		entry->start_bt = io->io_hdr.start_bt;
2025#endif
2026		bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2027		entry->cdb_len = io->scsiio.cdb_len;
2028		if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2029			entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2030
2031		if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2032			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2033
2034		if (io->io_hdr.flags & CTL_FLAG_ABORT)
2035			entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2036
2037		if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2038			entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2039
2040		if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2041			entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2042	}
2043
2044	return (retval);
2045}
2046
2047static void *
2048ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2049		 size_t error_str_len)
2050{
2051	void *kptr;
2052
2053	kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2054
2055	if (copyin(user_addr, kptr, len) != 0) {
2056		snprintf(error_str, error_str_len, "Error copying %d bytes "
2057			 "from user address %p to kernel address %p", len,
2058			 user_addr, kptr);
2059		free(kptr, M_CTL);
2060		return (NULL);
2061	}
2062
2063	return (kptr);
2064}
2065
2066static void
2067ctl_free_args(int num_be_args, struct ctl_be_arg *be_args)
2068{
2069	int i;
2070
2071	if (be_args == NULL)
2072		return;
2073
2074	for (i = 0; i < num_be_args; i++) {
2075		free(be_args[i].kname, M_CTL);
2076		free(be_args[i].kvalue, M_CTL);
2077	}
2078
2079	free(be_args, M_CTL);
2080}
2081
2082static struct ctl_be_arg *
2083ctl_copyin_args(int num_be_args, struct ctl_be_arg *be_args,
2084		char *error_str, size_t error_str_len)
2085{
2086	struct ctl_be_arg *args;
2087	int i;
2088
2089	args = ctl_copyin_alloc(be_args, num_be_args * sizeof(*be_args),
2090				error_str, error_str_len);
2091
2092	if (args == NULL)
2093		goto bailout;
2094
2095	for (i = 0; i < num_be_args; i++) {
2096		args[i].kname = NULL;
2097		args[i].kvalue = NULL;
2098	}
2099
2100	for (i = 0; i < num_be_args; i++) {
2101		uint8_t *tmpptr;
2102
2103		args[i].kname = ctl_copyin_alloc(args[i].name,
2104			args[i].namelen, error_str, error_str_len);
2105		if (args[i].kname == NULL)
2106			goto bailout;
2107
2108		if (args[i].kname[args[i].namelen - 1] != '\0') {
2109			snprintf(error_str, error_str_len, "Argument %d "
2110				 "name is not NUL-terminated", i);
2111			goto bailout;
2112		}
2113
2114		args[i].kvalue = NULL;
2115
2116		tmpptr = ctl_copyin_alloc(args[i].value,
2117			args[i].vallen, error_str, error_str_len);
2118		if (tmpptr == NULL)
2119			goto bailout;
2120
2121		args[i].kvalue = tmpptr;
2122
2123		if ((args[i].flags & CTL_BEARG_ASCII)
2124		 && (tmpptr[args[i].vallen - 1] != '\0')) {
2125			snprintf(error_str, error_str_len, "Argument %d "
2126				 "value is not NUL-terminated", i);
2127			goto bailout;
2128		}
2129	}
2130
2131	return (args);
2132bailout:
2133
2134	ctl_free_args(num_be_args, args);
2135
2136	return (NULL);
2137}
2138
2139/*
2140 * Escape characters that are illegal or not recommended in XML.
2141 */
2142int
2143ctl_sbuf_printf_esc(struct sbuf *sb, char *str)
2144{
2145	int retval;
2146
2147	retval = 0;
2148
2149	for (; *str; str++) {
2150		switch (*str) {
2151		case '&':
2152			retval = sbuf_printf(sb, "&amp;");
2153			break;
2154		case '>':
2155			retval = sbuf_printf(sb, "&gt;");
2156			break;
2157		case '<':
2158			retval = sbuf_printf(sb, "&lt;");
2159			break;
2160		default:
2161			retval = sbuf_putc(sb, *str);
2162			break;
2163		}
2164
2165		if (retval != 0)
2166			break;
2167
2168	}
2169
2170	return (retval);
2171}
2172
2173static int
2174ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2175	  struct thread *td)
2176{
2177	struct ctl_softc *softc;
2178	int retval;
2179
2180	softc = control_softc;
2181
2182	retval = 0;
2183
2184	switch (cmd) {
2185	case CTL_IO: {
2186		union ctl_io *io;
2187		void *pool_tmp;
2188
2189		/*
2190		 * If we haven't been "enabled", don't allow any SCSI I/O
2191		 * to this FETD.
2192		 */
2193		if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) {
2194			retval = EPERM;
2195			break;
2196		}
2197
2198		io = ctl_alloc_io(softc->ioctl_info.fe.ctl_pool_ref);
2199		if (io == NULL) {
2200			printf("ctl_ioctl: can't allocate ctl_io!\n");
2201			retval = ENOSPC;
2202			break;
2203		}
2204
2205		/*
2206		 * Need to save the pool reference so it doesn't get
2207		 * spammed by the user's ctl_io.
2208		 */
2209		pool_tmp = io->io_hdr.pool;
2210
2211		memcpy(io, (void *)addr, sizeof(*io));
2212
2213		io->io_hdr.pool = pool_tmp;
2214		/*
2215		 * No status yet, so make sure the status is set properly.
2216		 */
2217		io->io_hdr.status = CTL_STATUS_NONE;
2218
2219		/*
2220		 * The user sets the initiator ID, target and LUN IDs.
2221		 */
2222		io->io_hdr.nexus.targ_port = softc->ioctl_info.fe.targ_port;
2223		io->io_hdr.flags |= CTL_FLAG_USER_REQ;
2224		if ((io->io_hdr.io_type == CTL_IO_SCSI)
2225		 && (io->scsiio.tag_type != CTL_TAG_UNTAGGED))
2226			io->scsiio.tag_num = softc->ioctl_info.cur_tag_num++;
2227
2228		retval = ctl_ioctl_submit_wait(io);
2229
2230		if (retval != 0) {
2231			ctl_free_io(io);
2232			break;
2233		}
2234
2235		memcpy((void *)addr, io, sizeof(*io));
2236
2237		/* return this to our pool */
2238		ctl_free_io(io);
2239
2240		break;
2241	}
2242	case CTL_ENABLE_PORT:
2243	case CTL_DISABLE_PORT:
2244	case CTL_SET_PORT_WWNS: {
2245		struct ctl_frontend *fe;
2246		struct ctl_port_entry *entry;
2247
2248		entry = (struct ctl_port_entry *)addr;
2249
2250		mtx_lock(&softc->ctl_lock);
2251		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2252			int action, done;
2253
2254			action = 0;
2255			done = 0;
2256
2257			if ((entry->port_type == CTL_PORT_NONE)
2258			 && (entry->targ_port == fe->targ_port)) {
2259				/*
2260				 * If the user only wants to enable or
2261				 * disable or set WWNs on a specific port,
2262				 * do the operation and we're done.
2263				 */
2264				action = 1;
2265				done = 1;
2266			} else if (entry->port_type & fe->port_type) {
2267				/*
2268				 * Compare the user's type mask with the
2269				 * particular frontend type to see if we
2270				 * have a match.
2271				 */
2272				action = 1;
2273				done = 0;
2274
2275				/*
2276				 * Make sure the user isn't trying to set
2277				 * WWNs on multiple ports at the same time.
2278				 */
2279				if (cmd == CTL_SET_PORT_WWNS) {
2280					printf("%s: Can't set WWNs on "
2281					       "multiple ports\n", __func__);
2282					retval = EINVAL;
2283					break;
2284				}
2285			}
2286			if (action != 0) {
2287				/*
2288				 * XXX KDM we have to drop the lock here,
2289				 * because the online/offline operations
2290				 * can potentially block.  We need to
2291				 * reference count the frontends so they
2292				 * can't go away,
2293				 */
2294				mtx_unlock(&softc->ctl_lock);
2295
2296				if (cmd == CTL_ENABLE_PORT) {
2297					struct ctl_lun *lun;
2298
2299					STAILQ_FOREACH(lun, &softc->lun_list,
2300						       links) {
2301						fe->lun_enable(fe->targ_lun_arg,
2302						    lun->target,
2303						    lun->lun);
2304					}
2305
2306					ctl_frontend_online(fe);
2307				} else if (cmd == CTL_DISABLE_PORT) {
2308					struct ctl_lun *lun;
2309
2310					ctl_frontend_offline(fe);
2311
2312					STAILQ_FOREACH(lun, &softc->lun_list,
2313						       links) {
2314						fe->lun_disable(
2315						    fe->targ_lun_arg,
2316						    lun->target,
2317						    lun->lun);
2318					}
2319				}
2320
2321				mtx_lock(&softc->ctl_lock);
2322
2323				if (cmd == CTL_SET_PORT_WWNS)
2324					ctl_frontend_set_wwns(fe,
2325					    (entry->flags & CTL_PORT_WWNN_VALID) ?
2326					    1 : 0, entry->wwnn,
2327					    (entry->flags & CTL_PORT_WWPN_VALID) ?
2328					    1 : 0, entry->wwpn);
2329			}
2330			if (done != 0)
2331				break;
2332		}
2333		mtx_unlock(&softc->ctl_lock);
2334		break;
2335	}
2336	case CTL_GET_PORT_LIST: {
2337		struct ctl_frontend *fe;
2338		struct ctl_port_list *list;
2339		int i;
2340
2341		list = (struct ctl_port_list *)addr;
2342
2343		if (list->alloc_len != (list->alloc_num *
2344		    sizeof(struct ctl_port_entry))) {
2345			printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2346			       "alloc_num %u * sizeof(struct ctl_port_entry) "
2347			       "%zu\n", __func__, list->alloc_len,
2348			       list->alloc_num, sizeof(struct ctl_port_entry));
2349			retval = EINVAL;
2350			break;
2351		}
2352		list->fill_len = 0;
2353		list->fill_num = 0;
2354		list->dropped_num = 0;
2355		i = 0;
2356		mtx_lock(&softc->ctl_lock);
2357		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2358			struct ctl_port_entry entry, *list_entry;
2359
2360			if (list->fill_num >= list->alloc_num) {
2361				list->dropped_num++;
2362				continue;
2363			}
2364
2365			entry.port_type = fe->port_type;
2366			strlcpy(entry.port_name, fe->port_name,
2367				sizeof(entry.port_name));
2368			entry.targ_port = fe->targ_port;
2369			entry.physical_port = fe->physical_port;
2370			entry.virtual_port = fe->virtual_port;
2371			entry.wwnn = fe->wwnn;
2372			entry.wwpn = fe->wwpn;
2373			if (fe->status & CTL_PORT_STATUS_ONLINE)
2374				entry.online = 1;
2375			else
2376				entry.online = 0;
2377
2378			list_entry = &list->entries[i];
2379
2380			retval = copyout(&entry, list_entry, sizeof(entry));
2381			if (retval != 0) {
2382				printf("%s: CTL_GET_PORT_LIST: copyout "
2383				       "returned %d\n", __func__, retval);
2384				break;
2385			}
2386			i++;
2387			list->fill_num++;
2388			list->fill_len += sizeof(entry);
2389		}
2390		mtx_unlock(&softc->ctl_lock);
2391
2392		/*
2393		 * If this is non-zero, we had a copyout fault, so there's
2394		 * probably no point in attempting to set the status inside
2395		 * the structure.
2396		 */
2397		if (retval != 0)
2398			break;
2399
2400		if (list->dropped_num > 0)
2401			list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2402		else
2403			list->status = CTL_PORT_LIST_OK;
2404		break;
2405	}
2406	case CTL_DUMP_OOA: {
2407		struct ctl_lun *lun;
2408		union ctl_io *io;
2409		char printbuf[128];
2410		struct sbuf sb;
2411
2412		mtx_lock(&softc->ctl_lock);
2413		printf("Dumping OOA queues:\n");
2414		STAILQ_FOREACH(lun, &softc->lun_list, links) {
2415			for (io = (union ctl_io *)TAILQ_FIRST(
2416			     &lun->ooa_queue); io != NULL;
2417			     io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2418			     ooa_links)) {
2419				sbuf_new(&sb, printbuf, sizeof(printbuf),
2420					 SBUF_FIXEDLEN);
2421				sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2422					    (intmax_t)lun->lun,
2423					    io->scsiio.tag_num,
2424					    (io->io_hdr.flags &
2425					    CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2426					    (io->io_hdr.flags &
2427					    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2428					    (io->io_hdr.flags &
2429					    CTL_FLAG_ABORT) ? " ABORT" : "",
2430			                    (io->io_hdr.flags &
2431		                        CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2432				ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2433				sbuf_finish(&sb);
2434				printf("%s\n", sbuf_data(&sb));
2435			}
2436		}
2437		printf("OOA queues dump done\n");
2438		mtx_unlock(&softc->ctl_lock);
2439		break;
2440	}
2441	case CTL_GET_OOA: {
2442		struct ctl_lun *lun;
2443		struct ctl_ooa *ooa_hdr;
2444		struct ctl_ooa_entry *entries;
2445		uint32_t cur_fill_num;
2446
2447		ooa_hdr = (struct ctl_ooa *)addr;
2448
2449		if ((ooa_hdr->alloc_len == 0)
2450		 || (ooa_hdr->alloc_num == 0)) {
2451			printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2452			       "must be non-zero\n", __func__,
2453			       ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2454			retval = EINVAL;
2455			break;
2456		}
2457
2458		if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2459		    sizeof(struct ctl_ooa_entry))) {
2460			printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2461			       "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2462			       __func__, ooa_hdr->alloc_len,
2463			       ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2464			retval = EINVAL;
2465			break;
2466		}
2467
2468		entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2469		if (entries == NULL) {
2470			printf("%s: could not allocate %d bytes for OOA "
2471			       "dump\n", __func__, ooa_hdr->alloc_len);
2472			retval = ENOMEM;
2473			break;
2474		}
2475
2476		mtx_lock(&softc->ctl_lock);
2477		if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2478		 && ((ooa_hdr->lun_num > CTL_MAX_LUNS)
2479		  || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2480			mtx_unlock(&softc->ctl_lock);
2481			free(entries, M_CTL);
2482			printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2483			       __func__, (uintmax_t)ooa_hdr->lun_num);
2484			retval = EINVAL;
2485			break;
2486		}
2487
2488		cur_fill_num = 0;
2489
2490		if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2491			STAILQ_FOREACH(lun, &softc->lun_list, links) {
2492				retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2493					ooa_hdr, entries);
2494				if (retval != 0)
2495					break;
2496			}
2497			if (retval != 0) {
2498				mtx_unlock(&softc->ctl_lock);
2499				free(entries, M_CTL);
2500				break;
2501			}
2502		} else {
2503			lun = softc->ctl_luns[ooa_hdr->lun_num];
2504
2505			retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2506						    entries);
2507		}
2508		mtx_unlock(&softc->ctl_lock);
2509
2510		ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2511		ooa_hdr->fill_len = ooa_hdr->fill_num *
2512			sizeof(struct ctl_ooa_entry);
2513		retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2514		if (retval != 0) {
2515			printf("%s: error copying out %d bytes for OOA dump\n",
2516			       __func__, ooa_hdr->fill_len);
2517		}
2518
2519		getbintime(&ooa_hdr->cur_bt);
2520
2521		if (cur_fill_num > ooa_hdr->alloc_num) {
2522			ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2523			ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2524		} else {
2525			ooa_hdr->dropped_num = 0;
2526			ooa_hdr->status = CTL_OOA_OK;
2527		}
2528
2529		free(entries, M_CTL);
2530		break;
2531	}
2532	case CTL_CHECK_OOA: {
2533		union ctl_io *io;
2534		struct ctl_lun *lun;
2535		struct ctl_ooa_info *ooa_info;
2536
2537
2538		ooa_info = (struct ctl_ooa_info *)addr;
2539
2540		if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2541			ooa_info->status = CTL_OOA_INVALID_LUN;
2542			break;
2543		}
2544		mtx_lock(&softc->ctl_lock);
2545		lun = softc->ctl_luns[ooa_info->lun_id];
2546		if (lun == NULL) {
2547			mtx_unlock(&softc->ctl_lock);
2548			ooa_info->status = CTL_OOA_INVALID_LUN;
2549			break;
2550		}
2551
2552		ooa_info->num_entries = 0;
2553		for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2554		     io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2555		     &io->io_hdr, ooa_links)) {
2556			ooa_info->num_entries++;
2557		}
2558
2559		mtx_unlock(&softc->ctl_lock);
2560		ooa_info->status = CTL_OOA_SUCCESS;
2561
2562		break;
2563	}
2564	case CTL_HARD_START:
2565	case CTL_HARD_STOP: {
2566		struct ctl_fe_ioctl_startstop_info ss_info;
2567		struct cfi_metatask *metatask;
2568		struct mtx hs_mtx;
2569
2570		mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF);
2571
2572		cv_init(&ss_info.sem, "hard start/stop cv" );
2573
2574		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2575		if (metatask == NULL) {
2576			retval = ENOMEM;
2577			mtx_destroy(&hs_mtx);
2578			break;
2579		}
2580
2581		if (cmd == CTL_HARD_START)
2582			metatask->tasktype = CFI_TASK_STARTUP;
2583		else
2584			metatask->tasktype = CFI_TASK_SHUTDOWN;
2585
2586		metatask->callback = ctl_ioctl_hard_startstop_callback;
2587		metatask->callback_arg = &ss_info;
2588
2589		cfi_action(metatask);
2590
2591		/* Wait for the callback */
2592		mtx_lock(&hs_mtx);
2593		cv_wait_sig(&ss_info.sem, &hs_mtx);
2594		mtx_unlock(&hs_mtx);
2595
2596		/*
2597		 * All information has been copied from the metatask by the
2598		 * time cv_broadcast() is called, so we free the metatask here.
2599		 */
2600		cfi_free_metatask(metatask);
2601
2602		memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info));
2603
2604		mtx_destroy(&hs_mtx);
2605		break;
2606	}
2607	case CTL_BBRREAD: {
2608		struct ctl_bbrread_info *bbr_info;
2609		struct ctl_fe_ioctl_bbrread_info fe_bbr_info;
2610		struct mtx bbr_mtx;
2611		struct cfi_metatask *metatask;
2612
2613		bbr_info = (struct ctl_bbrread_info *)addr;
2614
2615		bzero(&fe_bbr_info, sizeof(fe_bbr_info));
2616
2617		bzero(&bbr_mtx, sizeof(bbr_mtx));
2618		mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF);
2619
2620		fe_bbr_info.bbr_info = bbr_info;
2621		fe_bbr_info.lock = &bbr_mtx;
2622
2623		cv_init(&fe_bbr_info.sem, "BBR read cv");
2624		metatask = cfi_alloc_metatask(/*can_wait*/ 1);
2625
2626		if (metatask == NULL) {
2627			mtx_destroy(&bbr_mtx);
2628			cv_destroy(&fe_bbr_info.sem);
2629			retval = ENOMEM;
2630			break;
2631		}
2632		metatask->tasktype = CFI_TASK_BBRREAD;
2633		metatask->callback = ctl_ioctl_bbrread_callback;
2634		metatask->callback_arg = &fe_bbr_info;
2635		metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num;
2636		metatask->taskinfo.bbrread.lba = bbr_info->lba;
2637		metatask->taskinfo.bbrread.len = bbr_info->len;
2638
2639		cfi_action(metatask);
2640
2641		mtx_lock(&bbr_mtx);
2642		while (fe_bbr_info.wakeup_done == 0)
2643			cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx);
2644		mtx_unlock(&bbr_mtx);
2645
2646		bbr_info->status = metatask->status;
2647		bbr_info->bbr_status = metatask->taskinfo.bbrread.status;
2648		bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status;
2649		memcpy(&bbr_info->sense_data,
2650		       &metatask->taskinfo.bbrread.sense_data,
2651		       ctl_min(sizeof(bbr_info->sense_data),
2652			       sizeof(metatask->taskinfo.bbrread.sense_data)));
2653
2654		cfi_free_metatask(metatask);
2655
2656		mtx_destroy(&bbr_mtx);
2657		cv_destroy(&fe_bbr_info.sem);
2658
2659		break;
2660	}
2661	case CTL_DELAY_IO: {
2662		struct ctl_io_delay_info *delay_info;
2663#ifdef CTL_IO_DELAY
2664		struct ctl_lun *lun;
2665#endif /* CTL_IO_DELAY */
2666
2667		delay_info = (struct ctl_io_delay_info *)addr;
2668
2669#ifdef CTL_IO_DELAY
2670		mtx_lock(&softc->ctl_lock);
2671
2672		if ((delay_info->lun_id > CTL_MAX_LUNS)
2673		 || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2674			delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2675		} else {
2676			lun = softc->ctl_luns[delay_info->lun_id];
2677
2678			delay_info->status = CTL_DELAY_STATUS_OK;
2679
2680			switch (delay_info->delay_type) {
2681			case CTL_DELAY_TYPE_CONT:
2682				break;
2683			case CTL_DELAY_TYPE_ONESHOT:
2684				break;
2685			default:
2686				delay_info->status =
2687					CTL_DELAY_STATUS_INVALID_TYPE;
2688				break;
2689			}
2690
2691			switch (delay_info->delay_loc) {
2692			case CTL_DELAY_LOC_DATAMOVE:
2693				lun->delay_info.datamove_type =
2694					delay_info->delay_type;
2695				lun->delay_info.datamove_delay =
2696					delay_info->delay_secs;
2697				break;
2698			case CTL_DELAY_LOC_DONE:
2699				lun->delay_info.done_type =
2700					delay_info->delay_type;
2701				lun->delay_info.done_delay =
2702					delay_info->delay_secs;
2703				break;
2704			default:
2705				delay_info->status =
2706					CTL_DELAY_STATUS_INVALID_LOC;
2707				break;
2708			}
2709		}
2710
2711		mtx_unlock(&softc->ctl_lock);
2712#else
2713		delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2714#endif /* CTL_IO_DELAY */
2715		break;
2716	}
2717	case CTL_REALSYNC_SET: {
2718		int *syncstate;
2719
2720		syncstate = (int *)addr;
2721
2722		mtx_lock(&softc->ctl_lock);
2723		switch (*syncstate) {
2724		case 0:
2725			softc->flags &= ~CTL_FLAG_REAL_SYNC;
2726			break;
2727		case 1:
2728			softc->flags |= CTL_FLAG_REAL_SYNC;
2729			break;
2730		default:
2731			retval = EINVAL;
2732			break;
2733		}
2734		mtx_unlock(&softc->ctl_lock);
2735		break;
2736	}
2737	case CTL_REALSYNC_GET: {
2738		int *syncstate;
2739
2740		syncstate = (int*)addr;
2741
2742		mtx_lock(&softc->ctl_lock);
2743		if (softc->flags & CTL_FLAG_REAL_SYNC)
2744			*syncstate = 1;
2745		else
2746			*syncstate = 0;
2747		mtx_unlock(&softc->ctl_lock);
2748
2749		break;
2750	}
2751	case CTL_SETSYNC:
2752	case CTL_GETSYNC: {
2753		struct ctl_sync_info *sync_info;
2754		struct ctl_lun *lun;
2755
2756		sync_info = (struct ctl_sync_info *)addr;
2757
2758		mtx_lock(&softc->ctl_lock);
2759		lun = softc->ctl_luns[sync_info->lun_id];
2760		if (lun == NULL) {
2761			mtx_unlock(&softc->ctl_lock);
2762			sync_info->status = CTL_GS_SYNC_NO_LUN;
2763		}
2764		/*
2765		 * Get or set the sync interval.  We're not bounds checking
2766		 * in the set case, hopefully the user won't do something
2767		 * silly.
2768		 */
2769		if (cmd == CTL_GETSYNC)
2770			sync_info->sync_interval = lun->sync_interval;
2771		else
2772			lun->sync_interval = sync_info->sync_interval;
2773
2774		mtx_unlock(&softc->ctl_lock);
2775
2776		sync_info->status = CTL_GS_SYNC_OK;
2777
2778		break;
2779	}
2780	case CTL_GETSTATS: {
2781		struct ctl_stats *stats;
2782		struct ctl_lun *lun;
2783		int i;
2784
2785		stats = (struct ctl_stats *)addr;
2786
2787		if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2788		     stats->alloc_len) {
2789			stats->status = CTL_SS_NEED_MORE_SPACE;
2790			stats->num_luns = softc->num_luns;
2791			break;
2792		}
2793		/*
2794		 * XXX KDM no locking here.  If the LUN list changes,
2795		 * things can blow up.
2796		 */
2797		for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2798		     i++, lun = STAILQ_NEXT(lun, links)) {
2799			retval = copyout(&lun->stats, &stats->lun_stats[i],
2800					 sizeof(lun->stats));
2801			if (retval != 0)
2802				break;
2803		}
2804		stats->num_luns = softc->num_luns;
2805		stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2806				 softc->num_luns;
2807		stats->status = CTL_SS_OK;
2808#ifdef CTL_TIME_IO
2809		stats->flags = CTL_STATS_FLAG_TIME_VALID;
2810#else
2811		stats->flags = CTL_STATS_FLAG_NONE;
2812#endif
2813		getnanouptime(&stats->timestamp);
2814		break;
2815	}
2816	case CTL_ERROR_INJECT: {
2817		struct ctl_error_desc *err_desc, *new_err_desc;
2818		struct ctl_lun *lun;
2819
2820		err_desc = (struct ctl_error_desc *)addr;
2821
2822		new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2823				      M_WAITOK | M_ZERO);
2824		bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2825
2826		mtx_lock(&softc->ctl_lock);
2827		lun = softc->ctl_luns[err_desc->lun_id];
2828		if (lun == NULL) {
2829			mtx_unlock(&softc->ctl_lock);
2830			printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2831			       __func__, (uintmax_t)err_desc->lun_id);
2832			retval = EINVAL;
2833			break;
2834		}
2835
2836		/*
2837		 * We could do some checking here to verify the validity
2838		 * of the request, but given the complexity of error
2839		 * injection requests, the checking logic would be fairly
2840		 * complex.
2841		 *
2842		 * For now, if the request is invalid, it just won't get
2843		 * executed and might get deleted.
2844		 */
2845		STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2846
2847		/*
2848		 * XXX KDM check to make sure the serial number is unique,
2849		 * in case we somehow manage to wrap.  That shouldn't
2850		 * happen for a very long time, but it's the right thing to
2851		 * do.
2852		 */
2853		new_err_desc->serial = lun->error_serial;
2854		err_desc->serial = lun->error_serial;
2855		lun->error_serial++;
2856
2857		mtx_unlock(&softc->ctl_lock);
2858		break;
2859	}
2860	case CTL_ERROR_INJECT_DELETE: {
2861		struct ctl_error_desc *delete_desc, *desc, *desc2;
2862		struct ctl_lun *lun;
2863		int delete_done;
2864
2865		delete_desc = (struct ctl_error_desc *)addr;
2866		delete_done = 0;
2867
2868		mtx_lock(&softc->ctl_lock);
2869		lun = softc->ctl_luns[delete_desc->lun_id];
2870		if (lun == NULL) {
2871			mtx_unlock(&softc->ctl_lock);
2872			printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2873			       __func__, (uintmax_t)delete_desc->lun_id);
2874			retval = EINVAL;
2875			break;
2876		}
2877		STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2878			if (desc->serial != delete_desc->serial)
2879				continue;
2880
2881			STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2882				      links);
2883			free(desc, M_CTL);
2884			delete_done = 1;
2885		}
2886		mtx_unlock(&softc->ctl_lock);
2887		if (delete_done == 0) {
2888			printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2889			       "error serial %ju on LUN %u\n", __func__,
2890			       delete_desc->serial, delete_desc->lun_id);
2891			retval = EINVAL;
2892			break;
2893		}
2894		break;
2895	}
2896	case CTL_DUMP_STRUCTS: {
2897		int i, j, k;
2898		struct ctl_frontend *fe;
2899
2900		printf("CTL IID to WWPN map start:\n");
2901		for (i = 0; i < CTL_MAX_PORTS; i++) {
2902			for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2903				if (softc->wwpn_iid[i][j].in_use == 0)
2904					continue;
2905
2906				printf("port %d iid %u WWPN %#jx\n",
2907				       softc->wwpn_iid[i][j].port,
2908				       softc->wwpn_iid[i][j].iid,
2909				       (uintmax_t)softc->wwpn_iid[i][j].wwpn);
2910			}
2911		}
2912		printf("CTL IID to WWPN map end\n");
2913		printf("CTL Persistent Reservation information start:\n");
2914		for (i = 0; i < CTL_MAX_LUNS; i++) {
2915			struct ctl_lun *lun;
2916
2917			lun = softc->ctl_luns[i];
2918
2919			if ((lun == NULL)
2920			 || ((lun->flags & CTL_LUN_DISABLED) != 0))
2921				continue;
2922
2923			for (j = 0; j < (CTL_MAX_PORTS * 2); j++) {
2924				for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2925					if (lun->per_res[j+k].registered == 0)
2926						continue;
2927					printf("LUN %d port %d iid %d key "
2928					       "%#jx\n", i, j, k,
2929					       (uintmax_t)scsi_8btou64(
2930					       lun->per_res[j+k].res_key.key));
2931				}
2932			}
2933		}
2934		printf("CTL Persistent Reservation information end\n");
2935		printf("CTL Frontends:\n");
2936		/*
2937		 * XXX KDM calling this without a lock.  We'd likely want
2938		 * to drop the lock before calling the frontend's dump
2939		 * routine anyway.
2940		 */
2941		STAILQ_FOREACH(fe, &softc->fe_list, links) {
2942			printf("Frontend %s Type %u pport %d vport %d WWNN "
2943			       "%#jx WWPN %#jx\n", fe->port_name, fe->port_type,
2944			       fe->physical_port, fe->virtual_port,
2945			       (uintmax_t)fe->wwnn, (uintmax_t)fe->wwpn);
2946
2947			/*
2948			 * Frontends are not required to support the dump
2949			 * routine.
2950			 */
2951			if (fe->fe_dump == NULL)
2952				continue;
2953
2954			fe->fe_dump();
2955		}
2956		printf("CTL Frontend information end\n");
2957		break;
2958	}
2959	case CTL_LUN_REQ: {
2960		struct ctl_lun_req *lun_req;
2961		struct ctl_backend_driver *backend;
2962
2963		lun_req = (struct ctl_lun_req *)addr;
2964
2965		backend = ctl_backend_find(lun_req->backend);
2966		if (backend == NULL) {
2967			lun_req->status = CTL_LUN_ERROR;
2968			snprintf(lun_req->error_str,
2969				 sizeof(lun_req->error_str),
2970				 "Backend \"%s\" not found.",
2971				 lun_req->backend);
2972			break;
2973		}
2974		if (lun_req->num_be_args > 0) {
2975			lun_req->kern_be_args = ctl_copyin_args(
2976				lun_req->num_be_args,
2977				lun_req->be_args,
2978				lun_req->error_str,
2979				sizeof(lun_req->error_str));
2980			if (lun_req->kern_be_args == NULL) {
2981				lun_req->status = CTL_LUN_ERROR;
2982				break;
2983			}
2984		}
2985
2986		retval = backend->ioctl(dev, cmd, addr, flag, td);
2987
2988		if (lun_req->num_be_args > 0) {
2989			ctl_free_args(lun_req->num_be_args,
2990				      lun_req->kern_be_args);
2991		}
2992		break;
2993	}
2994	case CTL_LUN_LIST: {
2995		struct sbuf *sb;
2996		struct ctl_lun *lun;
2997		struct ctl_lun_list *list;
2998		struct ctl_be_lun_option *opt;
2999
3000		list = (struct ctl_lun_list *)addr;
3001
3002		/*
3003		 * Allocate a fixed length sbuf here, based on the length
3004		 * of the user's buffer.  We could allocate an auto-extending
3005		 * buffer, and then tell the user how much larger our
3006		 * amount of data is than his buffer, but that presents
3007		 * some problems:
3008		 *
3009		 * 1.  The sbuf(9) routines use a blocking malloc, and so
3010		 *     we can't hold a lock while calling them with an
3011		 *     auto-extending buffer.
3012 		 *
3013		 * 2.  There is not currently a LUN reference counting
3014		 *     mechanism, outside of outstanding transactions on
3015		 *     the LUN's OOA queue.  So a LUN could go away on us
3016		 *     while we're getting the LUN number, backend-specific
3017		 *     information, etc.  Thus, given the way things
3018		 *     currently work, we need to hold the CTL lock while
3019		 *     grabbing LUN information.
3020		 *
3021		 * So, from the user's standpoint, the best thing to do is
3022		 * allocate what he thinks is a reasonable buffer length,
3023		 * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3024		 * double the buffer length and try again.  (And repeat
3025		 * that until he succeeds.)
3026		 */
3027		sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3028		if (sb == NULL) {
3029			list->status = CTL_LUN_LIST_ERROR;
3030			snprintf(list->error_str, sizeof(list->error_str),
3031				 "Unable to allocate %d bytes for LUN list",
3032				 list->alloc_len);
3033			break;
3034		}
3035
3036		sbuf_printf(sb, "<ctllunlist>\n");
3037
3038		mtx_lock(&softc->ctl_lock);
3039
3040		STAILQ_FOREACH(lun, &softc->lun_list, links) {
3041			retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3042					     (uintmax_t)lun->lun);
3043
3044			/*
3045			 * Bail out as soon as we see that we've overfilled
3046			 * the buffer.
3047			 */
3048			if (retval != 0)
3049				break;
3050
3051			retval = sbuf_printf(sb, "<backend_type>%s"
3052					     "</backend_type>\n",
3053					     (lun->backend == NULL) ?  "none" :
3054					     lun->backend->name);
3055
3056			if (retval != 0)
3057				break;
3058
3059			retval = sbuf_printf(sb, "<lun_type>%d</lun_type>\n",
3060					     lun->be_lun->lun_type);
3061
3062			if (retval != 0)
3063				break;
3064
3065			if (lun->backend == NULL) {
3066				retval = sbuf_printf(sb, "</lun>\n");
3067				if (retval != 0)
3068					break;
3069				continue;
3070			}
3071
3072			retval = sbuf_printf(sb, "<size>%ju</size>\n",
3073					     (lun->be_lun->maxlba > 0) ?
3074					     lun->be_lun->maxlba + 1 : 0);
3075
3076			if (retval != 0)
3077				break;
3078
3079			retval = sbuf_printf(sb, "<blocksize>%u</blocksize>\n",
3080					     lun->be_lun->blocksize);
3081
3082			if (retval != 0)
3083				break;
3084
3085			retval = sbuf_printf(sb, "<serial_number>");
3086
3087			if (retval != 0)
3088				break;
3089
3090			retval = ctl_sbuf_printf_esc(sb,
3091						     lun->be_lun->serial_num);
3092
3093			if (retval != 0)
3094				break;
3095
3096			retval = sbuf_printf(sb, "</serial_number>\n");
3097
3098			if (retval != 0)
3099				break;
3100
3101			retval = sbuf_printf(sb, "<device_id>");
3102
3103			if (retval != 0)
3104				break;
3105
3106			retval = ctl_sbuf_printf_esc(sb,lun->be_lun->device_id);
3107
3108			if (retval != 0)
3109				break;
3110
3111			retval = sbuf_printf(sb, "</device_id>\n");
3112
3113			if (retval != 0)
3114				break;
3115
3116			if (lun->backend->lun_info != NULL) {
3117				retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3118				if (retval != 0)
3119					break;
3120			}
3121			STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3122				retval = sbuf_printf(sb, "<%s>%s</%s>", opt->name, opt->value, opt->name);
3123				if (retval != 0)
3124					break;
3125			}
3126
3127			retval = sbuf_printf(sb, "</lun>\n");
3128
3129			if (retval != 0)
3130				break;
3131		}
3132		mtx_unlock(&softc->ctl_lock);
3133
3134		if ((retval != 0)
3135		 || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3136			retval = 0;
3137			sbuf_delete(sb);
3138			list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3139			snprintf(list->error_str, sizeof(list->error_str),
3140				 "Out of space, %d bytes is too small",
3141				 list->alloc_len);
3142			break;
3143		}
3144
3145		sbuf_finish(sb);
3146
3147		retval = copyout(sbuf_data(sb), list->lun_xml,
3148				 sbuf_len(sb) + 1);
3149
3150		list->fill_len = sbuf_len(sb) + 1;
3151		list->status = CTL_LUN_LIST_OK;
3152		sbuf_delete(sb);
3153		break;
3154	}
3155	case CTL_ISCSI: {
3156		struct ctl_iscsi *ci;
3157		struct ctl_frontend *fe;
3158
3159		ci = (struct ctl_iscsi *)addr;
3160
3161		mtx_lock(&softc->ctl_lock);
3162		STAILQ_FOREACH(fe, &softc->fe_list, links) {
3163			if (strcmp(fe->port_name, "iscsi") == 0)
3164				break;
3165		}
3166		mtx_unlock(&softc->ctl_lock);
3167
3168		if (fe == NULL) {
3169			ci->status = CTL_ISCSI_ERROR;
3170			snprintf(ci->error_str, sizeof(ci->error_str), "Backend \"iscsi\" not found.");
3171			break;
3172		}
3173
3174		retval = fe->ioctl(dev, cmd, addr, flag, td);
3175		break;
3176	}
3177	default: {
3178		/* XXX KDM should we fix this? */
3179#if 0
3180		struct ctl_backend_driver *backend;
3181		unsigned int type;
3182		int found;
3183
3184		found = 0;
3185
3186		/*
3187		 * We encode the backend type as the ioctl type for backend
3188		 * ioctls.  So parse it out here, and then search for a
3189		 * backend of this type.
3190		 */
3191		type = _IOC_TYPE(cmd);
3192
3193		STAILQ_FOREACH(backend, &softc->be_list, links) {
3194			if (backend->type == type) {
3195				found = 1;
3196				break;
3197			}
3198		}
3199		if (found == 0) {
3200			printf("ctl: unknown ioctl command %#lx or backend "
3201			       "%d\n", cmd, type);
3202			retval = EINVAL;
3203			break;
3204		}
3205		retval = backend->ioctl(dev, cmd, addr, flag, td);
3206#endif
3207		retval = ENOTTY;
3208		break;
3209	}
3210	}
3211	return (retval);
3212}
3213
3214uint32_t
3215ctl_get_initindex(struct ctl_nexus *nexus)
3216{
3217	if (nexus->targ_port < CTL_MAX_PORTS)
3218		return (nexus->initid.id +
3219			(nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3220	else
3221		return (nexus->initid.id +
3222		       ((nexus->targ_port - CTL_MAX_PORTS) *
3223			CTL_MAX_INIT_PER_PORT));
3224}
3225
3226uint32_t
3227ctl_get_resindex(struct ctl_nexus *nexus)
3228{
3229	return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3230}
3231
3232uint32_t
3233ctl_port_idx(int port_num)
3234{
3235	if (port_num < CTL_MAX_PORTS)
3236		return(port_num);
3237	else
3238		return(port_num - CTL_MAX_PORTS);
3239}
3240
3241/*
3242 * Note:  This only works for bitmask sizes that are at least 32 bits, and
3243 * that are a power of 2.
3244 */
3245int
3246ctl_ffz(uint32_t *mask, uint32_t size)
3247{
3248	uint32_t num_chunks, num_pieces;
3249	int i, j;
3250
3251	num_chunks = (size >> 5);
3252	if (num_chunks == 0)
3253		num_chunks++;
3254	num_pieces = ctl_min((sizeof(uint32_t) * 8), size);
3255
3256	for (i = 0; i < num_chunks; i++) {
3257		for (j = 0; j < num_pieces; j++) {
3258			if ((mask[i] & (1 << j)) == 0)
3259				return ((i << 5) + j);
3260		}
3261	}
3262
3263	return (-1);
3264}
3265
3266int
3267ctl_set_mask(uint32_t *mask, uint32_t bit)
3268{
3269	uint32_t chunk, piece;
3270
3271	chunk = bit >> 5;
3272	piece = bit % (sizeof(uint32_t) * 8);
3273
3274	if ((mask[chunk] & (1 << piece)) != 0)
3275		return (-1);
3276	else
3277		mask[chunk] |= (1 << piece);
3278
3279	return (0);
3280}
3281
3282int
3283ctl_clear_mask(uint32_t *mask, uint32_t bit)
3284{
3285	uint32_t chunk, piece;
3286
3287	chunk = bit >> 5;
3288	piece = bit % (sizeof(uint32_t) * 8);
3289
3290	if ((mask[chunk] & (1 << piece)) == 0)
3291		return (-1);
3292	else
3293		mask[chunk] &= ~(1 << piece);
3294
3295	return (0);
3296}
3297
3298int
3299ctl_is_set(uint32_t *mask, uint32_t bit)
3300{
3301	uint32_t chunk, piece;
3302
3303	chunk = bit >> 5;
3304	piece = bit % (sizeof(uint32_t) * 8);
3305
3306	if ((mask[chunk] & (1 << piece)) == 0)
3307		return (0);
3308	else
3309		return (1);
3310}
3311
3312#ifdef unused
3313/*
3314 * The bus, target and lun are optional, they can be filled in later.
3315 * can_wait is used to determine whether we can wait on the malloc or not.
3316 */
3317union ctl_io*
3318ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target,
3319	      uint32_t targ_lun, int can_wait)
3320{
3321	union ctl_io *io;
3322
3323	if (can_wait)
3324		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK);
3325	else
3326		io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3327
3328	if (io != NULL) {
3329		io->io_hdr.io_type = io_type;
3330		io->io_hdr.targ_port = targ_port;
3331		/*
3332		 * XXX KDM this needs to change/go away.  We need to move
3333		 * to a preallocated pool of ctl_scsiio structures.
3334		 */
3335		io->io_hdr.nexus.targ_target.id = targ_target;
3336		io->io_hdr.nexus.targ_lun = targ_lun;
3337	}
3338
3339	return (io);
3340}
3341
3342void
3343ctl_kfree_io(union ctl_io *io)
3344{
3345	free(io, M_CTL);
3346}
3347#endif /* unused */
3348
3349/*
3350 * ctl_softc, pool_type, total_ctl_io are passed in.
3351 * npool is passed out.
3352 */
3353int
3354ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type,
3355		uint32_t total_ctl_io, struct ctl_io_pool **npool)
3356{
3357	uint32_t i;
3358	union ctl_io *cur_io, *next_io;
3359	struct ctl_io_pool *pool;
3360	int retval;
3361
3362	retval = 0;
3363
3364	pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3365					    M_NOWAIT | M_ZERO);
3366	if (pool == NULL) {
3367		retval = ENOMEM;
3368		goto bailout;
3369	}
3370
3371	pool->type = pool_type;
3372	pool->ctl_softc = ctl_softc;
3373
3374	mtx_lock(&ctl_softc->pool_lock);
3375	pool->id = ctl_softc->cur_pool_id++;
3376	mtx_unlock(&ctl_softc->pool_lock);
3377
3378	pool->flags = CTL_POOL_FLAG_NONE;
3379	pool->refcount = 1;		/* Reference for validity. */
3380	STAILQ_INIT(&pool->free_queue);
3381
3382	/*
3383	 * XXX KDM other options here:
3384	 * - allocate a page at a time
3385	 * - allocate one big chunk of memory.
3386	 * Page allocation might work well, but would take a little more
3387	 * tracking.
3388	 */
3389	for (i = 0; i < total_ctl_io; i++) {
3390		cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTL,
3391						M_NOWAIT);
3392		if (cur_io == NULL) {
3393			retval = ENOMEM;
3394			break;
3395		}
3396		cur_io->io_hdr.pool = pool;
3397		STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links);
3398		pool->total_ctl_io++;
3399		pool->free_ctl_io++;
3400	}
3401
3402	if (retval != 0) {
3403		for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3404		     cur_io != NULL; cur_io = next_io) {
3405			next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr,
3406							      links);
3407			STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr,
3408				      ctl_io_hdr, links);
3409			free(cur_io, M_CTL);
3410		}
3411
3412		free(pool, M_CTL);
3413		goto bailout;
3414	}
3415	mtx_lock(&ctl_softc->pool_lock);
3416	ctl_softc->num_pools++;
3417	STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links);
3418	/*
3419	 * Increment our usage count if this is an external consumer, so we
3420	 * can't get unloaded until the external consumer (most likely a
3421	 * FETD) unloads and frees his pool.
3422	 *
3423	 * XXX KDM will this increment the caller's module use count, or
3424	 * mine?
3425	 */
3426#if 0
3427	if ((pool_type != CTL_POOL_EMERGENCY)
3428	 && (pool_type != CTL_POOL_INTERNAL)
3429	 && (pool_type != CTL_POOL_IOCTL)
3430	 && (pool_type != CTL_POOL_4OTHERSC))
3431		MOD_INC_USE_COUNT;
3432#endif
3433
3434	mtx_unlock(&ctl_softc->pool_lock);
3435
3436	*npool = pool;
3437
3438bailout:
3439
3440	return (retval);
3441}
3442
3443static int
3444ctl_pool_acquire(struct ctl_io_pool *pool)
3445{
3446
3447	mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED);
3448
3449	if (pool->flags & CTL_POOL_FLAG_INVALID)
3450		return (EINVAL);
3451
3452	pool->refcount++;
3453
3454	return (0);
3455}
3456
3457static void
3458ctl_pool_release(struct ctl_io_pool *pool)
3459{
3460	struct ctl_softc *ctl_softc = pool->ctl_softc;
3461	union ctl_io *io;
3462
3463	mtx_assert(&ctl_softc->pool_lock, MA_OWNED);
3464
3465	if (--pool->refcount != 0)
3466		return;
3467
3468	while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) {
3469		STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr,
3470			      links);
3471		free(io, M_CTL);
3472	}
3473
3474	STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links);
3475	ctl_softc->num_pools--;
3476
3477	/*
3478	 * XXX KDM will this decrement the caller's usage count or mine?
3479	 */
3480#if 0
3481	if ((pool->type != CTL_POOL_EMERGENCY)
3482	 && (pool->type != CTL_POOL_INTERNAL)
3483	 && (pool->type != CTL_POOL_IOCTL))
3484		MOD_DEC_USE_COUNT;
3485#endif
3486
3487	free(pool, M_CTL);
3488}
3489
3490void
3491ctl_pool_free(struct ctl_io_pool *pool)
3492{
3493	struct ctl_softc *ctl_softc;
3494
3495	if (pool == NULL)
3496		return;
3497
3498	ctl_softc = pool->ctl_softc;
3499	mtx_lock(&ctl_softc->pool_lock);
3500	pool->flags |= CTL_POOL_FLAG_INVALID;
3501	ctl_pool_release(pool);
3502	mtx_unlock(&ctl_softc->pool_lock);
3503}
3504
3505/*
3506 * This routine does not block (except for spinlocks of course).
3507 * It tries to allocate a ctl_io union from the caller's pool as quickly as
3508 * possible.
3509 */
3510union ctl_io *
3511ctl_alloc_io(void *pool_ref)
3512{
3513	union ctl_io *io;
3514	struct ctl_softc *ctl_softc;
3515	struct ctl_io_pool *pool, *npool;
3516	struct ctl_io_pool *emergency_pool;
3517
3518	pool = (struct ctl_io_pool *)pool_ref;
3519
3520	if (pool == NULL) {
3521		printf("%s: pool is NULL\n", __func__);
3522		return (NULL);
3523	}
3524
3525	emergency_pool = NULL;
3526
3527	ctl_softc = pool->ctl_softc;
3528
3529	mtx_lock(&ctl_softc->pool_lock);
3530	/*
3531	 * First, try to get the io structure from the user's pool.
3532	 */
3533	if (ctl_pool_acquire(pool) == 0) {
3534		io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue);
3535		if (io != NULL) {
3536			STAILQ_REMOVE_HEAD(&pool->free_queue, links);
3537			pool->total_allocated++;
3538			pool->free_ctl_io--;
3539			mtx_unlock(&ctl_softc->pool_lock);
3540			return (io);
3541		} else
3542			ctl_pool_release(pool);
3543	}
3544	/*
3545	 * If he doesn't have any io structures left, search for an
3546	 * emergency pool and grab one from there.
3547	 */
3548	STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) {
3549		if (npool->type != CTL_POOL_EMERGENCY)
3550			continue;
3551
3552		if (ctl_pool_acquire(npool) != 0)
3553			continue;
3554
3555		emergency_pool = npool;
3556
3557		io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue);
3558		if (io != NULL) {
3559			STAILQ_REMOVE_HEAD(&npool->free_queue, links);
3560			npool->total_allocated++;
3561			npool->free_ctl_io--;
3562			mtx_unlock(&ctl_softc->pool_lock);
3563			return (io);
3564		} else
3565			ctl_pool_release(npool);
3566	}
3567
3568	/* Drop the spinlock before we malloc */
3569	mtx_unlock(&ctl_softc->pool_lock);
3570
3571	/*
3572	 * The emergency pool (if it exists) didn't have one, so try an
3573	 * atomic (i.e. nonblocking) malloc and see if we get lucky.
3574	 */
3575	io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT);
3576	if (io != NULL) {
3577		/*
3578		 * If the emergency pool exists but is empty, add this
3579		 * ctl_io to its list when it gets freed.
3580		 */
3581		if (emergency_pool != NULL) {
3582			mtx_lock(&ctl_softc->pool_lock);
3583			if (ctl_pool_acquire(emergency_pool) == 0) {
3584				io->io_hdr.pool = emergency_pool;
3585				emergency_pool->total_ctl_io++;
3586				/*
3587				 * Need to bump this, otherwise
3588				 * total_allocated and total_freed won't
3589				 * match when we no longer have anything
3590				 * outstanding.
3591				 */
3592				emergency_pool->total_allocated++;
3593			}
3594			mtx_unlock(&ctl_softc->pool_lock);
3595		} else
3596			io->io_hdr.pool = NULL;
3597	}
3598
3599	return (io);
3600}
3601
3602void
3603ctl_free_io(union ctl_io *io)
3604{
3605	if (io == NULL)
3606		return;
3607
3608	/*
3609	 * If this ctl_io has a pool, return it to that pool.
3610	 */
3611	if (io->io_hdr.pool != NULL) {
3612		struct ctl_io_pool *pool;
3613#if 0
3614		struct ctl_softc *ctl_softc;
3615		union ctl_io *tmp_io;
3616		unsigned long xflags;
3617		int i;
3618
3619		ctl_softc = control_softc;
3620#endif
3621
3622		pool = (struct ctl_io_pool *)io->io_hdr.pool;
3623
3624		mtx_lock(&pool->ctl_softc->pool_lock);
3625#if 0
3626		save_flags(xflags);
3627
3628		for (i = 0, tmp_io = (union ctl_io *)STAILQ_FIRST(
3629		     &ctl_softc->task_queue); tmp_io != NULL; i++,
3630		     tmp_io = (union ctl_io *)STAILQ_NEXT(&tmp_io->io_hdr,
3631		     links)) {
3632			if (tmp_io == io) {
3633				printf("%s: %p is still on the task queue!\n",
3634				       __func__, tmp_io);
3635				printf("%s: (%d): type %d "
3636				       "msg %d cdb %x iptl: "
3637				       "%d:%d:%d:%d tag 0x%04x "
3638				       "flg %#lx\n",
3639					__func__, i,
3640					tmp_io->io_hdr.io_type,
3641					tmp_io->io_hdr.msg_type,
3642					tmp_io->scsiio.cdb[0],
3643					tmp_io->io_hdr.nexus.initid.id,
3644					tmp_io->io_hdr.nexus.targ_port,
3645					tmp_io->io_hdr.nexus.targ_target.id,
3646					tmp_io->io_hdr.nexus.targ_lun,
3647					(tmp_io->io_hdr.io_type ==
3648					CTL_IO_TASK) ?
3649					tmp_io->taskio.tag_num :
3650					tmp_io->scsiio.tag_num,
3651					xflags);
3652				panic("I/O still on the task queue!");
3653			}
3654		}
3655#endif
3656		io->io_hdr.io_type = 0xff;
3657		STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links);
3658		pool->total_freed++;
3659		pool->free_ctl_io++;
3660		ctl_pool_release(pool);
3661		mtx_unlock(&pool->ctl_softc->pool_lock);
3662	} else {
3663		/*
3664		 * Otherwise, just free it.  We probably malloced it and
3665		 * the emergency pool wasn't available.
3666		 */
3667		free(io, M_CTL);
3668	}
3669
3670}
3671
3672void
3673ctl_zero_io(union ctl_io *io)
3674{
3675	void *pool_ref;
3676
3677	if (io == NULL)
3678		return;
3679
3680	/*
3681	 * May need to preserve linked list pointers at some point too.
3682	 */
3683	pool_ref = io->io_hdr.pool;
3684
3685	memset(io, 0, sizeof(*io));
3686
3687	io->io_hdr.pool = pool_ref;
3688}
3689
3690/*
3691 * This routine is currently used for internal copies of ctl_ios that need
3692 * to persist for some reason after we've already returned status to the
3693 * FETD.  (Thus the flag set.)
3694 *
3695 * XXX XXX
3696 * Note that this makes a blind copy of all fields in the ctl_io, except
3697 * for the pool reference.  This includes any memory that has been
3698 * allocated!  That memory will no longer be valid after done has been
3699 * called, so this would be VERY DANGEROUS for command that actually does
3700 * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3701 * start and stop commands, which don't transfer any data, so this is not a
3702 * problem.  If it is used for anything else, the caller would also need to
3703 * allocate data buffer space and this routine would need to be modified to
3704 * copy the data buffer(s) as well.
3705 */
3706void
3707ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3708{
3709	void *pool_ref;
3710
3711	if ((src == NULL)
3712	 || (dest == NULL))
3713		return;
3714
3715	/*
3716	 * May need to preserve linked list pointers at some point too.
3717	 */
3718	pool_ref = dest->io_hdr.pool;
3719
3720	memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest)));
3721
3722	dest->io_hdr.pool = pool_ref;
3723	/*
3724	 * We need to know that this is an internal copy, and doesn't need
3725	 * to get passed back to the FETD that allocated it.
3726	 */
3727	dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3728}
3729
3730#ifdef NEEDTOPORT
3731static void
3732ctl_update_power_subpage(struct copan_power_subpage *page)
3733{
3734	int num_luns, num_partitions, config_type;
3735	struct ctl_softc *softc;
3736	cs_BOOL_t aor_present, shelf_50pct_power;
3737	cs_raidset_personality_t rs_type;
3738	int max_active_luns;
3739
3740	softc = control_softc;
3741
3742	/* subtract out the processor LUN */
3743	num_luns = softc->num_luns - 1;
3744	/*
3745	 * Default to 7 LUNs active, which was the only number we allowed
3746	 * in the past.
3747	 */
3748	max_active_luns = 7;
3749
3750	num_partitions = config_GetRsPartitionInfo();
3751	config_type = config_GetConfigType();
3752	shelf_50pct_power = config_GetShelfPowerMode();
3753	aor_present = config_IsAorRsPresent();
3754
3755	rs_type = ddb_GetRsRaidType(1);
3756	if ((rs_type != CS_RAIDSET_PERSONALITY_RAID5)
3757	 && (rs_type != CS_RAIDSET_PERSONALITY_RAID1)) {
3758		EPRINT(0, "Unsupported RS type %d!", rs_type);
3759	}
3760
3761
3762	page->total_luns = num_luns;
3763
3764	switch (config_type) {
3765	case 40:
3766		/*
3767		 * In a 40 drive configuration, it doesn't matter what DC
3768		 * cards we have, whether we have AOR enabled or not,
3769		 * partitioning or not, or what type of RAIDset we have.
3770		 * In that scenario, we can power up every LUN we present
3771		 * to the user.
3772		 */
3773		max_active_luns = num_luns;
3774
3775		break;
3776	case 64:
3777		if (shelf_50pct_power == CS_FALSE) {
3778			/* 25% power */
3779			if (aor_present == CS_TRUE) {
3780				if (rs_type ==
3781				     CS_RAIDSET_PERSONALITY_RAID5) {
3782					max_active_luns = 7;
3783				} else if (rs_type ==
3784					 CS_RAIDSET_PERSONALITY_RAID1){
3785					max_active_luns = 14;
3786				} else {
3787					/* XXX KDM now what?? */
3788				}
3789			} else {
3790				if (rs_type ==
3791				     CS_RAIDSET_PERSONALITY_RAID5) {
3792					max_active_luns = 8;
3793				} else if (rs_type ==
3794					 CS_RAIDSET_PERSONALITY_RAID1){
3795					max_active_luns = 16;
3796				} else {
3797					/* XXX KDM now what?? */
3798				}
3799			}
3800		} else {
3801			/* 50% power */
3802			/*
3803			 * With 50% power in a 64 drive configuration, we
3804			 * can power all LUNs we present.
3805			 */
3806			max_active_luns = num_luns;
3807		}
3808		break;
3809	case 112:
3810		if (shelf_50pct_power == CS_FALSE) {
3811			/* 25% power */
3812			if (aor_present == CS_TRUE) {
3813				if (rs_type ==
3814				     CS_RAIDSET_PERSONALITY_RAID5) {
3815					max_active_luns = 7;
3816				} else if (rs_type ==
3817					 CS_RAIDSET_PERSONALITY_RAID1){
3818					max_active_luns = 14;
3819				} else {
3820					/* XXX KDM now what?? */
3821				}
3822			} else {
3823				if (rs_type ==
3824				     CS_RAIDSET_PERSONALITY_RAID5) {
3825					max_active_luns = 8;
3826				} else if (rs_type ==
3827					 CS_RAIDSET_PERSONALITY_RAID1){
3828					max_active_luns = 16;
3829				} else {
3830					/* XXX KDM now what?? */
3831				}
3832			}
3833		} else {
3834			/* 50% power */
3835			if (aor_present == CS_TRUE) {
3836				if (rs_type ==
3837				     CS_RAIDSET_PERSONALITY_RAID5) {
3838					max_active_luns = 14;
3839				} else if (rs_type ==
3840					 CS_RAIDSET_PERSONALITY_RAID1){
3841					/*
3842					 * We're assuming here that disk
3843					 * caching is enabled, and so we're
3844					 * able to power up half of each
3845					 * LUN, and cache all writes.
3846					 */
3847					max_active_luns = num_luns;
3848				} else {
3849					/* XXX KDM now what?? */
3850				}
3851			} else {
3852				if (rs_type ==
3853				     CS_RAIDSET_PERSONALITY_RAID5) {
3854					max_active_luns = 15;
3855				} else if (rs_type ==
3856					 CS_RAIDSET_PERSONALITY_RAID1){
3857					max_active_luns = 30;
3858				} else {
3859					/* XXX KDM now what?? */
3860				}
3861			}
3862		}
3863		break;
3864	default:
3865		/*
3866		 * In this case, we have an unknown configuration, so we
3867		 * just use the default from above.
3868		 */
3869		break;
3870	}
3871
3872	page->max_active_luns = max_active_luns;
3873#if 0
3874	printk("%s: total_luns = %d, max_active_luns = %d\n", __func__,
3875	       page->total_luns, page->max_active_luns);
3876#endif
3877}
3878#endif /* NEEDTOPORT */
3879
3880/*
3881 * This routine could be used in the future to load default and/or saved
3882 * mode page parameters for a particuar lun.
3883 */
3884static int
3885ctl_init_page_index(struct ctl_lun *lun)
3886{
3887	int i;
3888	struct ctl_page_index *page_index;
3889	struct ctl_softc *softc;
3890
3891	memcpy(&lun->mode_pages.index, page_index_template,
3892	       sizeof(page_index_template));
3893
3894	softc = lun->ctl_softc;
3895
3896	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3897
3898		page_index = &lun->mode_pages.index[i];
3899		/*
3900		 * If this is a disk-only mode page, there's no point in
3901		 * setting it up.  For some pages, we have to have some
3902		 * basic information about the disk in order to calculate the
3903		 * mode page data.
3904		 */
3905		if ((lun->be_lun->lun_type != T_DIRECT)
3906		 && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3907			continue;
3908
3909		switch (page_index->page_code & SMPH_PC_MASK) {
3910		case SMS_FORMAT_DEVICE_PAGE: {
3911			struct scsi_format_page *format_page;
3912
3913			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3914				panic("subpage is incorrect!");
3915
3916			/*
3917			 * Sectors per track are set above.  Bytes per
3918			 * sector need to be set here on a per-LUN basis.
3919			 */
3920			memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3921			       &format_page_default,
3922			       sizeof(format_page_default));
3923			memcpy(&lun->mode_pages.format_page[
3924			       CTL_PAGE_CHANGEABLE], &format_page_changeable,
3925			       sizeof(format_page_changeable));
3926			memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3927			       &format_page_default,
3928			       sizeof(format_page_default));
3929			memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3930			       &format_page_default,
3931			       sizeof(format_page_default));
3932
3933			format_page = &lun->mode_pages.format_page[
3934				CTL_PAGE_CURRENT];
3935			scsi_ulto2b(lun->be_lun->blocksize,
3936				    format_page->bytes_per_sector);
3937
3938			format_page = &lun->mode_pages.format_page[
3939				CTL_PAGE_DEFAULT];
3940			scsi_ulto2b(lun->be_lun->blocksize,
3941				    format_page->bytes_per_sector);
3942
3943			format_page = &lun->mode_pages.format_page[
3944				CTL_PAGE_SAVED];
3945			scsi_ulto2b(lun->be_lun->blocksize,
3946				    format_page->bytes_per_sector);
3947
3948			page_index->page_data =
3949				(uint8_t *)lun->mode_pages.format_page;
3950			break;
3951		}
3952		case SMS_RIGID_DISK_PAGE: {
3953			struct scsi_rigid_disk_page *rigid_disk_page;
3954			uint32_t sectors_per_cylinder;
3955			uint64_t cylinders;
3956#ifndef	__XSCALE__
3957			int shift;
3958#endif /* !__XSCALE__ */
3959
3960			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3961				panic("invalid subpage value %d",
3962				      page_index->subpage);
3963
3964			/*
3965			 * Rotation rate and sectors per track are set
3966			 * above.  We calculate the cylinders here based on
3967			 * capacity.  Due to the number of heads and
3968			 * sectors per track we're using, smaller arrays
3969			 * may turn out to have 0 cylinders.  Linux and
3970			 * FreeBSD don't pay attention to these mode pages
3971			 * to figure out capacity, but Solaris does.  It
3972			 * seems to deal with 0 cylinders just fine, and
3973			 * works out a fake geometry based on the capacity.
3974			 */
3975			memcpy(&lun->mode_pages.rigid_disk_page[
3976			       CTL_PAGE_CURRENT], &rigid_disk_page_default,
3977			       sizeof(rigid_disk_page_default));
3978			memcpy(&lun->mode_pages.rigid_disk_page[
3979			       CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3980			       sizeof(rigid_disk_page_changeable));
3981			memcpy(&lun->mode_pages.rigid_disk_page[
3982			       CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3983			       sizeof(rigid_disk_page_default));
3984			memcpy(&lun->mode_pages.rigid_disk_page[
3985			       CTL_PAGE_SAVED], &rigid_disk_page_default,
3986			       sizeof(rigid_disk_page_default));
3987
3988			sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3989				CTL_DEFAULT_HEADS;
3990
3991			/*
3992			 * The divide method here will be more accurate,
3993			 * probably, but results in floating point being
3994			 * used in the kernel on i386 (__udivdi3()).  On the
3995			 * XScale, though, __udivdi3() is implemented in
3996			 * software.
3997			 *
3998			 * The shift method for cylinder calculation is
3999			 * accurate if sectors_per_cylinder is a power of
4000			 * 2.  Otherwise it might be slightly off -- you
4001			 * might have a bit of a truncation problem.
4002			 */
4003#ifdef	__XSCALE__
4004			cylinders = (lun->be_lun->maxlba + 1) /
4005				sectors_per_cylinder;
4006#else
4007			for (shift = 31; shift > 0; shift--) {
4008				if (sectors_per_cylinder & (1 << shift))
4009					break;
4010			}
4011			cylinders = (lun->be_lun->maxlba + 1) >> shift;
4012#endif
4013
4014			/*
4015			 * We've basically got 3 bytes, or 24 bits for the
4016			 * cylinder size in the mode page.  If we're over,
4017			 * just round down to 2^24.
4018			 */
4019			if (cylinders > 0xffffff)
4020				cylinders = 0xffffff;
4021
4022			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4023				CTL_PAGE_CURRENT];
4024			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4025
4026			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4027				CTL_PAGE_DEFAULT];
4028			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4029
4030			rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4031				CTL_PAGE_SAVED];
4032			scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4033
4034			page_index->page_data =
4035				(uint8_t *)lun->mode_pages.rigid_disk_page;
4036			break;
4037		}
4038		case SMS_CACHING_PAGE: {
4039
4040			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4041				panic("invalid subpage value %d",
4042				      page_index->subpage);
4043			/*
4044			 * Defaults should be okay here, no calculations
4045			 * needed.
4046			 */
4047			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4048			       &caching_page_default,
4049			       sizeof(caching_page_default));
4050			memcpy(&lun->mode_pages.caching_page[
4051			       CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4052			       sizeof(caching_page_changeable));
4053			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4054			       &caching_page_default,
4055			       sizeof(caching_page_default));
4056			memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4057			       &caching_page_default,
4058			       sizeof(caching_page_default));
4059			page_index->page_data =
4060				(uint8_t *)lun->mode_pages.caching_page;
4061			break;
4062		}
4063		case SMS_CONTROL_MODE_PAGE: {
4064
4065			if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
4066				panic("invalid subpage value %d",
4067				      page_index->subpage);
4068
4069			/*
4070			 * Defaults should be okay here, no calculations
4071			 * needed.
4072			 */
4073			memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
4074			       &control_page_default,
4075			       sizeof(control_page_default));
4076			memcpy(&lun->mode_pages.control_page[
4077			       CTL_PAGE_CHANGEABLE], &control_page_changeable,
4078			       sizeof(control_page_changeable));
4079			memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
4080			       &control_page_default,
4081			       sizeof(control_page_default));
4082			memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
4083			       &control_page_default,
4084			       sizeof(control_page_default));
4085			page_index->page_data =
4086				(uint8_t *)lun->mode_pages.control_page;
4087			break;
4088
4089		}
4090		case SMS_VENDOR_SPECIFIC_PAGE:{
4091			switch (page_index->subpage) {
4092			case PWR_SUBPAGE_CODE: {
4093				struct copan_power_subpage *current_page,
4094							   *saved_page;
4095
4096				memcpy(&lun->mode_pages.power_subpage[
4097				       CTL_PAGE_CURRENT],
4098				       &power_page_default,
4099				       sizeof(power_page_default));
4100				memcpy(&lun->mode_pages.power_subpage[
4101				       CTL_PAGE_CHANGEABLE],
4102				       &power_page_changeable,
4103				       sizeof(power_page_changeable));
4104				memcpy(&lun->mode_pages.power_subpage[
4105				       CTL_PAGE_DEFAULT],
4106				       &power_page_default,
4107				       sizeof(power_page_default));
4108				memcpy(&lun->mode_pages.power_subpage[
4109				       CTL_PAGE_SAVED],
4110				       &power_page_default,
4111				       sizeof(power_page_default));
4112				page_index->page_data =
4113				    (uint8_t *)lun->mode_pages.power_subpage;
4114
4115				current_page = (struct copan_power_subpage *)
4116					(page_index->page_data +
4117					 (page_index->page_len *
4118					  CTL_PAGE_CURRENT));
4119			        saved_page = (struct copan_power_subpage *)
4120				        (page_index->page_data +
4121					 (page_index->page_len *
4122					  CTL_PAGE_SAVED));
4123				break;
4124			}
4125			case APS_SUBPAGE_CODE: {
4126				struct copan_aps_subpage *current_page,
4127							 *saved_page;
4128
4129				// This gets set multiple times but
4130				// it should always be the same. It's
4131				// only done during init so who cares.
4132				index_to_aps_page = i;
4133
4134				memcpy(&lun->mode_pages.aps_subpage[
4135				       CTL_PAGE_CURRENT],
4136				       &aps_page_default,
4137				       sizeof(aps_page_default));
4138				memcpy(&lun->mode_pages.aps_subpage[
4139				       CTL_PAGE_CHANGEABLE],
4140				       &aps_page_changeable,
4141				       sizeof(aps_page_changeable));
4142				memcpy(&lun->mode_pages.aps_subpage[
4143				       CTL_PAGE_DEFAULT],
4144				       &aps_page_default,
4145				       sizeof(aps_page_default));
4146				memcpy(&lun->mode_pages.aps_subpage[
4147				       CTL_PAGE_SAVED],
4148				       &aps_page_default,
4149				       sizeof(aps_page_default));
4150				page_index->page_data =
4151					(uint8_t *)lun->mode_pages.aps_subpage;
4152
4153				current_page = (struct copan_aps_subpage *)
4154					(page_index->page_data +
4155					 (page_index->page_len *
4156					  CTL_PAGE_CURRENT));
4157				saved_page = (struct copan_aps_subpage *)
4158					(page_index->page_data +
4159					 (page_index->page_len *
4160					  CTL_PAGE_SAVED));
4161				break;
4162			}
4163			case DBGCNF_SUBPAGE_CODE: {
4164				struct copan_debugconf_subpage *current_page,
4165							       *saved_page;
4166
4167				memcpy(&lun->mode_pages.debugconf_subpage[
4168				       CTL_PAGE_CURRENT],
4169				       &debugconf_page_default,
4170				       sizeof(debugconf_page_default));
4171				memcpy(&lun->mode_pages.debugconf_subpage[
4172				       CTL_PAGE_CHANGEABLE],
4173				       &debugconf_page_changeable,
4174				       sizeof(debugconf_page_changeable));
4175				memcpy(&lun->mode_pages.debugconf_subpage[
4176				       CTL_PAGE_DEFAULT],
4177				       &debugconf_page_default,
4178				       sizeof(debugconf_page_default));
4179				memcpy(&lun->mode_pages.debugconf_subpage[
4180				       CTL_PAGE_SAVED],
4181				       &debugconf_page_default,
4182				       sizeof(debugconf_page_default));
4183				page_index->page_data =
4184					(uint8_t *)lun->mode_pages.debugconf_subpage;
4185
4186				current_page = (struct copan_debugconf_subpage *)
4187					(page_index->page_data +
4188					 (page_index->page_len *
4189					  CTL_PAGE_CURRENT));
4190				saved_page = (struct copan_debugconf_subpage *)
4191					(page_index->page_data +
4192					 (page_index->page_len *
4193					  CTL_PAGE_SAVED));
4194				break;
4195			}
4196			default:
4197				panic("invalid subpage value %d",
4198				      page_index->subpage);
4199				break;
4200			}
4201   			break;
4202		}
4203		default:
4204			panic("invalid page value %d",
4205			      page_index->page_code & SMPH_PC_MASK);
4206			break;
4207    	}
4208	}
4209
4210	return (CTL_RETVAL_COMPLETE);
4211}
4212
4213/*
4214 * LUN allocation.
4215 *
4216 * Requirements:
4217 * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4218 *   wants us to allocate the LUN and he can block.
4219 * - ctl_softc is always set
4220 * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4221 *
4222 * Returns 0 for success, non-zero (errno) for failure.
4223 */
4224static int
4225ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4226	      struct ctl_be_lun *const be_lun, struct ctl_id target_id)
4227{
4228	struct ctl_lun *nlun, *lun;
4229	struct ctl_frontend *fe;
4230	int lun_number, i, lun_malloced;
4231
4232	if (be_lun == NULL)
4233		return (EINVAL);
4234
4235	/*
4236	 * We currently only support Direct Access or Processor LUN types.
4237	 */
4238	switch (be_lun->lun_type) {
4239	case T_DIRECT:
4240		break;
4241	case T_PROCESSOR:
4242		break;
4243	case T_SEQUENTIAL:
4244	case T_CHANGER:
4245	default:
4246		be_lun->lun_config_status(be_lun->be_lun,
4247					  CTL_LUN_CONFIG_FAILURE);
4248		break;
4249	}
4250	if (ctl_lun == NULL) {
4251		lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4252		lun_malloced = 1;
4253	} else {
4254		lun_malloced = 0;
4255		lun = ctl_lun;
4256	}
4257
4258	memset(lun, 0, sizeof(*lun));
4259	if (lun_malloced)
4260		lun->flags = CTL_LUN_MALLOCED;
4261
4262	mtx_lock(&ctl_softc->ctl_lock);
4263	/*
4264	 * See if the caller requested a particular LUN number.  If so, see
4265	 * if it is available.  Otherwise, allocate the first available LUN.
4266	 */
4267	if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4268		if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4269		 || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4270			mtx_unlock(&ctl_softc->ctl_lock);
4271			if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4272				printf("ctl: requested LUN ID %d is higher "
4273				       "than CTL_MAX_LUNS - 1 (%d)\n",
4274				       be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4275			} else {
4276				/*
4277				 * XXX KDM return an error, or just assign
4278				 * another LUN ID in this case??
4279				 */
4280				printf("ctl: requested LUN ID %d is already "
4281				       "in use\n", be_lun->req_lun_id);
4282			}
4283			if (lun->flags & CTL_LUN_MALLOCED)
4284				free(lun, M_CTL);
4285			be_lun->lun_config_status(be_lun->be_lun,
4286						  CTL_LUN_CONFIG_FAILURE);
4287			return (ENOSPC);
4288		}
4289		lun_number = be_lun->req_lun_id;
4290	} else {
4291		lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, CTL_MAX_LUNS);
4292		if (lun_number == -1) {
4293			mtx_unlock(&ctl_softc->ctl_lock);
4294			printf("ctl: can't allocate LUN on target %ju, out of "
4295			       "LUNs\n", (uintmax_t)target_id.id);
4296			if (lun->flags & CTL_LUN_MALLOCED)
4297				free(lun, M_CTL);
4298			be_lun->lun_config_status(be_lun->be_lun,
4299						  CTL_LUN_CONFIG_FAILURE);
4300			return (ENOSPC);
4301		}
4302	}
4303	ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4304
4305	lun->target = target_id;
4306	lun->lun = lun_number;
4307	lun->be_lun = be_lun;
4308	/*
4309	 * The processor LUN is always enabled.  Disk LUNs come on line
4310	 * disabled, and must be enabled by the backend.
4311	 */
4312	lun->flags |= CTL_LUN_DISABLED;
4313	lun->backend = be_lun->be;
4314	be_lun->ctl_lun = lun;
4315	be_lun->lun_id = lun_number;
4316	atomic_add_int(&be_lun->be->num_luns, 1);
4317	if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4318		lun->flags |= CTL_LUN_STOPPED;
4319
4320	if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4321		lun->flags |= CTL_LUN_INOPERABLE;
4322
4323	if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4324		lun->flags |= CTL_LUN_PRIMARY_SC;
4325
4326	lun->ctl_softc = ctl_softc;
4327	TAILQ_INIT(&lun->ooa_queue);
4328	TAILQ_INIT(&lun->blocked_queue);
4329	STAILQ_INIT(&lun->error_list);
4330
4331	/*
4332	 * Initialize the mode page index.
4333	 */
4334	ctl_init_page_index(lun);
4335
4336	/*
4337	 * Set the poweron UA for all initiators on this LUN only.
4338	 */
4339	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4340		lun->pending_sense[i].ua_pending = CTL_UA_POWERON;
4341
4342	/*
4343	 * Now, before we insert this lun on the lun list, set the lun
4344	 * inventory changed UA for all other luns.
4345	 */
4346	STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4347		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4348			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4349		}
4350	}
4351
4352	STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4353
4354	ctl_softc->ctl_luns[lun_number] = lun;
4355
4356	ctl_softc->num_luns++;
4357
4358	/* Setup statistics gathering */
4359	lun->stats.device_type = be_lun->lun_type;
4360	lun->stats.lun_number = lun_number;
4361	if (lun->stats.device_type == T_DIRECT)
4362		lun->stats.blocksize = be_lun->blocksize;
4363	else
4364		lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4365	for (i = 0;i < CTL_MAX_PORTS;i++)
4366		lun->stats.ports[i].targ_port = i;
4367
4368	mtx_unlock(&ctl_softc->ctl_lock);
4369
4370	lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4371
4372	/*
4373	 * Run through each registered FETD and bring it online if it isn't
4374	 * already.  Enable the target ID if it hasn't been enabled, and
4375	 * enable this particular LUN.
4376	 */
4377	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4378		int retval;
4379
4380		/*
4381		 * XXX KDM this only works for ONE TARGET ID.  We'll need
4382		 * to do things differently if we go to a multiple target
4383		 * ID scheme.
4384		 */
4385		if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) == 0) {
4386
4387			retval = fe->targ_enable(fe->targ_lun_arg, target_id);
4388			if (retval != 0) {
4389				printf("ctl_alloc_lun: FETD %s port %d "
4390				       "returned error %d for targ_enable on "
4391				       "target %ju\n", fe->port_name,
4392				       fe->targ_port, retval,
4393				       (uintmax_t)target_id.id);
4394			} else
4395				fe->status |= CTL_PORT_STATUS_TARG_ONLINE;
4396		}
4397
4398		retval = fe->lun_enable(fe->targ_lun_arg, target_id,lun_number);
4399		if (retval != 0) {
4400			printf("ctl_alloc_lun: FETD %s port %d returned error "
4401			       "%d for lun_enable on target %ju lun %d\n",
4402			       fe->port_name, fe->targ_port, retval,
4403			       (uintmax_t)target_id.id, lun_number);
4404		} else
4405			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4406	}
4407	return (0);
4408}
4409
4410/*
4411 * Delete a LUN.
4412 * Assumptions:
4413 * - LUN has already been marked invalid and any pending I/O has been taken
4414 *   care of.
4415 */
4416static int
4417ctl_free_lun(struct ctl_lun *lun)
4418{
4419	struct ctl_softc *softc;
4420#if 0
4421	struct ctl_frontend *fe;
4422#endif
4423	struct ctl_lun *nlun;
4424	union ctl_io *io, *next_io;
4425	int i;
4426
4427	softc = lun->ctl_softc;
4428
4429	mtx_assert(&softc->ctl_lock, MA_OWNED);
4430
4431	STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4432
4433	ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4434
4435	softc->ctl_luns[lun->lun] = NULL;
4436
4437	if (TAILQ_FIRST(&lun->ooa_queue) != NULL) {
4438		printf("ctl_free_lun: aieee!! freeing a LUN with "
4439		       "outstanding I/O!!\n");
4440	}
4441
4442	/*
4443	 * If we have anything pending on the RtR queue, remove it.
4444	 */
4445	for (io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue); io != NULL;
4446	     io = next_io) {
4447		uint32_t targ_lun;
4448
4449		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
4450		targ_lun = io->io_hdr.nexus.targ_lun;
4451		if (io->io_hdr.nexus.lun_map_fn != NULL)
4452			targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
4453		if ((io->io_hdr.nexus.targ_target.id == lun->target.id)
4454		 && (targ_lun == lun->lun))
4455			STAILQ_REMOVE(&softc->rtr_queue, &io->io_hdr,
4456				      ctl_io_hdr, links);
4457	}
4458
4459	/*
4460	 * Then remove everything from the blocked queue.
4461	 */
4462	for (io = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); io != NULL;
4463	     io = next_io) {
4464		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,blocked_links);
4465		TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr, blocked_links);
4466		io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
4467	}
4468
4469	/*
4470	 * Now clear out the OOA queue, and free all the I/O.
4471	 * XXX KDM should we notify the FETD here?  We probably need to
4472	 * quiesce the LUN before deleting it.
4473	 */
4474	for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL;
4475	     io = next_io) {
4476		next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links);
4477		TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
4478		ctl_free_io(io);
4479	}
4480
4481	softc->num_luns--;
4482
4483	/*
4484	 * XXX KDM this scheme only works for a single target/multiple LUN
4485	 * setup.  It needs to be revamped for a multiple target scheme.
4486	 *
4487	 * XXX KDM this results in fe->lun_disable() getting called twice,
4488	 * once when ctl_disable_lun() is called, and a second time here.
4489	 * We really need to re-think the LUN disable semantics.  There
4490	 * should probably be several steps/levels to LUN removal:
4491	 *  - disable
4492	 *  - invalidate
4493	 *  - free
4494 	 *
4495	 * Right now we only have a disable method when communicating to
4496	 * the front end ports, at least for individual LUNs.
4497	 */
4498#if 0
4499	STAILQ_FOREACH(fe, &softc->fe_list, links) {
4500		int retval;
4501
4502		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4503					 lun->lun);
4504		if (retval != 0) {
4505			printf("ctl_free_lun: FETD %s port %d returned error "
4506			       "%d for lun_disable on target %ju lun %jd\n",
4507			       fe->port_name, fe->targ_port, retval,
4508			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4509		}
4510
4511		if (STAILQ_FIRST(&softc->lun_list) == NULL) {
4512			fe->status &= ~CTL_PORT_STATUS_LUN_ONLINE;
4513
4514			retval = fe->targ_disable(fe->targ_lun_arg,lun->target);
4515			if (retval != 0) {
4516				printf("ctl_free_lun: FETD %s port %d "
4517				       "returned error %d for targ_disable on "
4518				       "target %ju\n", fe->port_name,
4519				       fe->targ_port, retval,
4520				       (uintmax_t)lun->target.id);
4521			} else
4522				fe->status &= ~CTL_PORT_STATUS_TARG_ONLINE;
4523
4524			if ((fe->status & CTL_PORT_STATUS_TARG_ONLINE) != 0)
4525				continue;
4526
4527#if 0
4528			fe->port_offline(fe->onoff_arg);
4529			fe->status &= ~CTL_PORT_STATUS_ONLINE;
4530#endif
4531		}
4532	}
4533#endif
4534
4535	/*
4536	 * Tell the backend to free resources, if this LUN has a backend.
4537	 */
4538	atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4539	lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4540
4541	if (lun->flags & CTL_LUN_MALLOCED)
4542		free(lun, M_CTL);
4543
4544	STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4545		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
4546			nlun->pending_sense[i].ua_pending |= CTL_UA_LUN_CHANGE;
4547		}
4548	}
4549
4550	return (0);
4551}
4552
4553static void
4554ctl_create_lun(struct ctl_be_lun *be_lun)
4555{
4556	struct ctl_softc *ctl_softc;
4557
4558	ctl_softc = control_softc;
4559
4560	/*
4561	 * ctl_alloc_lun() should handle all potential failure cases.
4562	 */
4563	ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target);
4564}
4565
4566int
4567ctl_add_lun(struct ctl_be_lun *be_lun)
4568{
4569	struct ctl_softc *ctl_softc;
4570
4571	ctl_softc = control_softc;
4572
4573	mtx_lock(&ctl_softc->ctl_lock);
4574	STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links);
4575	mtx_unlock(&ctl_softc->ctl_lock);
4576
4577	ctl_wakeup_thread();
4578
4579	return (0);
4580}
4581
4582int
4583ctl_enable_lun(struct ctl_be_lun *be_lun)
4584{
4585	struct ctl_softc *ctl_softc;
4586	struct ctl_frontend *fe, *nfe;
4587	struct ctl_lun *lun;
4588	int retval;
4589
4590	ctl_softc = control_softc;
4591
4592	lun = (struct ctl_lun *)be_lun->ctl_lun;
4593
4594	mtx_lock(&ctl_softc->ctl_lock);
4595	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4596		/*
4597		 * eh?  Why did we get called if the LUN is already
4598		 * enabled?
4599		 */
4600		mtx_unlock(&ctl_softc->ctl_lock);
4601		return (0);
4602	}
4603	lun->flags &= ~CTL_LUN_DISABLED;
4604
4605	for (fe = STAILQ_FIRST(&ctl_softc->fe_list); fe != NULL; fe = nfe) {
4606		nfe = STAILQ_NEXT(fe, links);
4607
4608		/*
4609		 * Drop the lock while we call the FETD's enable routine.
4610		 * This can lead to a callback into CTL (at least in the
4611		 * case of the internal initiator frontend.
4612		 */
4613		mtx_unlock(&ctl_softc->ctl_lock);
4614		retval = fe->lun_enable(fe->targ_lun_arg, lun->target,lun->lun);
4615		mtx_lock(&ctl_softc->ctl_lock);
4616		if (retval != 0) {
4617			printf("%s: FETD %s port %d returned error "
4618			       "%d for lun_enable on target %ju lun %jd\n",
4619			       __func__, fe->port_name, fe->targ_port, retval,
4620			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4621		}
4622#if 0
4623		 else {
4624            /* NOTE:  TODO:  why does lun enable affect port status? */
4625			fe->status |= CTL_PORT_STATUS_LUN_ONLINE;
4626		}
4627#endif
4628	}
4629
4630	mtx_unlock(&ctl_softc->ctl_lock);
4631
4632	return (0);
4633}
4634
4635int
4636ctl_disable_lun(struct ctl_be_lun *be_lun)
4637{
4638	struct ctl_softc *ctl_softc;
4639	struct ctl_frontend *fe;
4640	struct ctl_lun *lun;
4641	int retval;
4642
4643	ctl_softc = control_softc;
4644
4645	lun = (struct ctl_lun *)be_lun->ctl_lun;
4646
4647	mtx_lock(&ctl_softc->ctl_lock);
4648
4649	if (lun->flags & CTL_LUN_DISABLED) {
4650		mtx_unlock(&ctl_softc->ctl_lock);
4651		return (0);
4652	}
4653	lun->flags |= CTL_LUN_DISABLED;
4654
4655	STAILQ_FOREACH(fe, &ctl_softc->fe_list, links) {
4656		mtx_unlock(&ctl_softc->ctl_lock);
4657		/*
4658		 * Drop the lock before we call the frontend's disable
4659		 * routine, to avoid lock order reversals.
4660		 *
4661		 * XXX KDM what happens if the frontend list changes while
4662		 * we're traversing it?  It's unlikely, but should be handled.
4663		 */
4664		retval = fe->lun_disable(fe->targ_lun_arg, lun->target,
4665					 lun->lun);
4666		mtx_lock(&ctl_softc->ctl_lock);
4667		if (retval != 0) {
4668			printf("ctl_alloc_lun: FETD %s port %d returned error "
4669			       "%d for lun_disable on target %ju lun %jd\n",
4670			       fe->port_name, fe->targ_port, retval,
4671			       (uintmax_t)lun->target.id, (intmax_t)lun->lun);
4672		}
4673	}
4674
4675	mtx_unlock(&ctl_softc->ctl_lock);
4676
4677	return (0);
4678}
4679
4680int
4681ctl_start_lun(struct ctl_be_lun *be_lun)
4682{
4683	struct ctl_softc *ctl_softc;
4684	struct ctl_lun *lun;
4685
4686	ctl_softc = control_softc;
4687
4688	lun = (struct ctl_lun *)be_lun->ctl_lun;
4689
4690	mtx_lock(&ctl_softc->ctl_lock);
4691	lun->flags &= ~CTL_LUN_STOPPED;
4692	mtx_unlock(&ctl_softc->ctl_lock);
4693
4694	return (0);
4695}
4696
4697int
4698ctl_stop_lun(struct ctl_be_lun *be_lun)
4699{
4700	struct ctl_softc *ctl_softc;
4701	struct ctl_lun *lun;
4702
4703	ctl_softc = control_softc;
4704
4705	lun = (struct ctl_lun *)be_lun->ctl_lun;
4706
4707	mtx_lock(&ctl_softc->ctl_lock);
4708	lun->flags |= CTL_LUN_STOPPED;
4709	mtx_unlock(&ctl_softc->ctl_lock);
4710
4711	return (0);
4712}
4713
4714int
4715ctl_lun_offline(struct ctl_be_lun *be_lun)
4716{
4717	struct ctl_softc *ctl_softc;
4718	struct ctl_lun *lun;
4719
4720	ctl_softc = control_softc;
4721
4722	lun = (struct ctl_lun *)be_lun->ctl_lun;
4723
4724	mtx_lock(&ctl_softc->ctl_lock);
4725	lun->flags |= CTL_LUN_OFFLINE;
4726	mtx_unlock(&ctl_softc->ctl_lock);
4727
4728	return (0);
4729}
4730
4731int
4732ctl_lun_online(struct ctl_be_lun *be_lun)
4733{
4734	struct ctl_softc *ctl_softc;
4735	struct ctl_lun *lun;
4736
4737	ctl_softc = control_softc;
4738
4739	lun = (struct ctl_lun *)be_lun->ctl_lun;
4740
4741	mtx_lock(&ctl_softc->ctl_lock);
4742	lun->flags &= ~CTL_LUN_OFFLINE;
4743	mtx_unlock(&ctl_softc->ctl_lock);
4744
4745	return (0);
4746}
4747
4748int
4749ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4750{
4751	struct ctl_softc *ctl_softc;
4752	struct ctl_lun *lun;
4753
4754	ctl_softc = control_softc;
4755
4756	lun = (struct ctl_lun *)be_lun->ctl_lun;
4757
4758	mtx_lock(&ctl_softc->ctl_lock);
4759
4760	/*
4761	 * The LUN needs to be disabled before it can be marked invalid.
4762	 */
4763	if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4764		mtx_unlock(&ctl_softc->ctl_lock);
4765		return (-1);
4766	}
4767	/*
4768	 * Mark the LUN invalid.
4769	 */
4770	lun->flags |= CTL_LUN_INVALID;
4771
4772	/*
4773	 * If there is nothing in the OOA queue, go ahead and free the LUN.
4774	 * If we have something in the OOA queue, we'll free it when the
4775	 * last I/O completes.
4776	 */
4777	if (TAILQ_FIRST(&lun->ooa_queue) == NULL)
4778		ctl_free_lun(lun);
4779	mtx_unlock(&ctl_softc->ctl_lock);
4780
4781	return (0);
4782}
4783
4784int
4785ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4786{
4787	struct ctl_softc *ctl_softc;
4788	struct ctl_lun *lun;
4789
4790	ctl_softc = control_softc;
4791	lun = (struct ctl_lun *)be_lun->ctl_lun;
4792
4793	mtx_lock(&ctl_softc->ctl_lock);
4794	lun->flags |= CTL_LUN_INOPERABLE;
4795	mtx_unlock(&ctl_softc->ctl_lock);
4796
4797	return (0);
4798}
4799
4800int
4801ctl_lun_operable(struct ctl_be_lun *be_lun)
4802{
4803	struct ctl_softc *ctl_softc;
4804	struct ctl_lun *lun;
4805
4806	ctl_softc = control_softc;
4807	lun = (struct ctl_lun *)be_lun->ctl_lun;
4808
4809	mtx_lock(&ctl_softc->ctl_lock);
4810	lun->flags &= ~CTL_LUN_INOPERABLE;
4811	mtx_unlock(&ctl_softc->ctl_lock);
4812
4813	return (0);
4814}
4815
4816int
4817ctl_lun_power_lock(struct ctl_be_lun *be_lun, struct ctl_nexus *nexus,
4818		   int lock)
4819{
4820	struct ctl_softc *softc;
4821	struct ctl_lun *lun;
4822	struct copan_aps_subpage *current_sp;
4823	struct ctl_page_index *page_index;
4824	int i;
4825
4826	softc = control_softc;
4827
4828	mtx_lock(&softc->ctl_lock);
4829
4830	lun = (struct ctl_lun *)be_lun->ctl_lun;
4831
4832	page_index = NULL;
4833	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4834		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
4835		     APS_PAGE_CODE)
4836			continue;
4837
4838		if (lun->mode_pages.index[i].subpage != APS_SUBPAGE_CODE)
4839			continue;
4840		page_index = &lun->mode_pages.index[i];
4841	}
4842
4843	if (page_index == NULL) {
4844		mtx_unlock(&softc->ctl_lock);
4845		printf("%s: APS subpage not found for lun %ju!\n", __func__,
4846		       (uintmax_t)lun->lun);
4847		return (1);
4848	}
4849#if 0
4850	if ((softc->aps_locked_lun != 0)
4851	 && (softc->aps_locked_lun != lun->lun)) {
4852		printf("%s: attempt to lock LUN %llu when %llu is already "
4853		       "locked\n");
4854		mtx_unlock(&softc->ctl_lock);
4855		return (1);
4856	}
4857#endif
4858
4859	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
4860		(page_index->page_len * CTL_PAGE_CURRENT));
4861
4862	if (lock != 0) {
4863		current_sp->lock_active = APS_LOCK_ACTIVE;
4864		softc->aps_locked_lun = lun->lun;
4865	} else {
4866		current_sp->lock_active = 0;
4867		softc->aps_locked_lun = 0;
4868	}
4869
4870
4871	/*
4872	 * If we're in HA mode, try to send the lock message to the other
4873	 * side.
4874	 */
4875	if (ctl_is_single == 0) {
4876		int isc_retval;
4877		union ctl_ha_msg lock_msg;
4878
4879		lock_msg.hdr.nexus = *nexus;
4880		lock_msg.hdr.msg_type = CTL_MSG_APS_LOCK;
4881		if (lock != 0)
4882			lock_msg.aps.lock_flag = 1;
4883		else
4884			lock_msg.aps.lock_flag = 0;
4885		isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &lock_msg,
4886					 sizeof(lock_msg), 0);
4887		if (isc_retval > CTL_HA_STATUS_SUCCESS) {
4888			printf("%s: APS (lock=%d) error returned from "
4889			       "ctl_ha_msg_send: %d\n", __func__, lock, isc_retval);
4890			mtx_unlock(&softc->ctl_lock);
4891			return (1);
4892		}
4893	}
4894
4895	mtx_unlock(&softc->ctl_lock);
4896
4897	return (0);
4898}
4899
4900void
4901ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4902{
4903	struct ctl_lun *lun;
4904	struct ctl_softc *softc;
4905	int i;
4906
4907	softc = control_softc;
4908
4909	mtx_lock(&softc->ctl_lock);
4910
4911	lun = (struct ctl_lun *)be_lun->ctl_lun;
4912
4913	for (i = 0; i < CTL_MAX_INITIATORS; i++)
4914		lun->pending_sense[i].ua_pending |= CTL_UA_CAPACITY_CHANGED;
4915
4916	mtx_unlock(&softc->ctl_lock);
4917}
4918
4919/*
4920 * Backend "memory move is complete" callback for requests that never
4921 * make it down to say RAIDCore's configuration code.
4922 */
4923int
4924ctl_config_move_done(union ctl_io *io)
4925{
4926	int retval;
4927
4928	retval = CTL_RETVAL_COMPLETE;
4929
4930
4931	CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4932	/*
4933	 * XXX KDM this shouldn't happen, but what if it does?
4934	 */
4935	if (io->io_hdr.io_type != CTL_IO_SCSI)
4936		panic("I/O type isn't CTL_IO_SCSI!");
4937
4938	if ((io->io_hdr.port_status == 0)
4939	 && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4940	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE))
4941		io->io_hdr.status = CTL_SUCCESS;
4942	else if ((io->io_hdr.port_status != 0)
4943	      && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0)
4944	      && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){
4945		/*
4946		 * For hardware error sense keys, the sense key
4947		 * specific value is defined to be a retry count,
4948		 * but we use it to pass back an internal FETD
4949		 * error code.  XXX KDM  Hopefully the FETD is only
4950		 * using 16 bits for an error code, since that's
4951		 * all the space we have in the sks field.
4952		 */
4953		ctl_set_internal_failure(&io->scsiio,
4954					 /*sks_valid*/ 1,
4955					 /*retry_count*/
4956					 io->io_hdr.port_status);
4957		free(io->scsiio.kern_data_ptr, M_CTL);
4958		ctl_done(io);
4959		goto bailout;
4960	}
4961
4962	if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
4963	 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
4964	 || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4965		/*
4966		 * XXX KDM just assuming a single pointer here, and not a
4967		 * S/G list.  If we start using S/G lists for config data,
4968		 * we'll need to know how to clean them up here as well.
4969		 */
4970		free(io->scsiio.kern_data_ptr, M_CTL);
4971		/* Hopefully the user has already set the status... */
4972		ctl_done(io);
4973	} else {
4974		/*
4975		 * XXX KDM now we need to continue data movement.  Some
4976		 * options:
4977		 * - call ctl_scsiio() again?  We don't do this for data
4978		 *   writes, because for those at least we know ahead of
4979		 *   time where the write will go and how long it is.  For
4980		 *   config writes, though, that information is largely
4981		 *   contained within the write itself, thus we need to
4982		 *   parse out the data again.
4983		 *
4984		 * - Call some other function once the data is in?
4985		 */
4986
4987		/*
4988		 * XXX KDM call ctl_scsiio() again for now, and check flag
4989		 * bits to see whether we're allocated or not.
4990		 */
4991		retval = ctl_scsiio(&io->scsiio);
4992	}
4993bailout:
4994	return (retval);
4995}
4996
4997/*
4998 * This gets called by a backend driver when it is done with a
4999 * data_submit method.
5000 */
5001void
5002ctl_data_submit_done(union ctl_io *io)
5003{
5004	/*
5005	 * If the IO_CONT flag is set, we need to call the supplied
5006	 * function to continue processing the I/O, instead of completing
5007	 * the I/O just yet.
5008	 *
5009	 * If there is an error, though, we don't want to keep processing.
5010	 * Instead, just send status back to the initiator.
5011	 */
5012	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5013	    (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5014	    ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5015	     (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5016		io->scsiio.io_cont(io);
5017		return;
5018	}
5019	ctl_done(io);
5020}
5021
5022/*
5023 * This gets called by a backend driver when it is done with a
5024 * configuration write.
5025 */
5026void
5027ctl_config_write_done(union ctl_io *io)
5028{
5029	/*
5030	 * If the IO_CONT flag is set, we need to call the supplied
5031	 * function to continue processing the I/O, instead of completing
5032	 * the I/O just yet.
5033	 *
5034	 * If there is an error, though, we don't want to keep processing.
5035	 * Instead, just send status back to the initiator.
5036	 */
5037	if ((io->io_hdr.flags & CTL_FLAG_IO_CONT)
5038	 && (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)
5039	  || ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))) {
5040		io->scsiio.io_cont(io);
5041		return;
5042	}
5043	/*
5044	 * Since a configuration write can be done for commands that actually
5045	 * have data allocated, like write buffer, and commands that have
5046	 * no data, like start/stop unit, we need to check here.
5047	 */
5048	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
5049		free(io->scsiio.kern_data_ptr, M_CTL);
5050	ctl_done(io);
5051}
5052
5053/*
5054 * SCSI release command.
5055 */
5056int
5057ctl_scsi_release(struct ctl_scsiio *ctsio)
5058{
5059	int length, longid, thirdparty_id, resv_id;
5060	struct ctl_softc *ctl_softc;
5061	struct ctl_lun *lun;
5062
5063	length = 0;
5064	resv_id = 0;
5065
5066	CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5067
5068	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5069	ctl_softc = control_softc;
5070
5071	switch (ctsio->cdb[0]) {
5072	case RELEASE: {
5073		struct scsi_release *cdb;
5074
5075		cdb = (struct scsi_release *)ctsio->cdb;
5076		if ((cdb->byte2 & 0x1f) != 0) {
5077			ctl_set_invalid_field(ctsio,
5078					      /*sks_valid*/ 1,
5079					      /*command*/ 1,
5080					      /*field*/ 1,
5081					      /*bit_valid*/ 0,
5082					      /*bit*/ 0);
5083			ctl_done((union ctl_io *)ctsio);
5084			return (CTL_RETVAL_COMPLETE);
5085		}
5086		break;
5087	}
5088	case RELEASE_10: {
5089		struct scsi_release_10 *cdb;
5090
5091		cdb = (struct scsi_release_10 *)ctsio->cdb;
5092
5093		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5094			ctl_set_invalid_field(ctsio,
5095					      /*sks_valid*/ 1,
5096					      /*command*/ 1,
5097					      /*field*/ 1,
5098					      /*bit_valid*/ 1,
5099					      /*bit*/ 0);
5100			ctl_done((union ctl_io *)ctsio);
5101			return (CTL_RETVAL_COMPLETE);
5102
5103		}
5104
5105		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5106			ctl_set_invalid_field(ctsio,
5107					      /*sks_valid*/ 1,
5108					      /*command*/ 1,
5109					      /*field*/ 1,
5110					      /*bit_valid*/ 1,
5111					      /*bit*/ 4);
5112			ctl_done((union ctl_io *)ctsio);
5113			return (CTL_RETVAL_COMPLETE);
5114		}
5115
5116		if (cdb->byte2 & SR10_LONGID)
5117			longid = 1;
5118		else
5119			thirdparty_id = cdb->thirdparty_id;
5120
5121		resv_id = cdb->resv_id;
5122		length = scsi_2btoul(cdb->length);
5123		break;
5124	}
5125	}
5126
5127
5128	/*
5129	 * XXX KDM right now, we only support LUN reservation.  We don't
5130	 * support 3rd party reservations, or extent reservations, which
5131	 * might actually need the parameter list.  If we've gotten this
5132	 * far, we've got a LUN reservation.  Anything else got kicked out
5133	 * above.  So, according to SPC, ignore the length.
5134	 */
5135	length = 0;
5136
5137	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5138	 && (length > 0)) {
5139		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5140		ctsio->kern_data_len = length;
5141		ctsio->kern_total_len = length;
5142		ctsio->kern_data_resid = 0;
5143		ctsio->kern_rel_offset = 0;
5144		ctsio->kern_sg_entries = 0;
5145		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5146		ctsio->be_move_done = ctl_config_move_done;
5147		ctl_datamove((union ctl_io *)ctsio);
5148
5149		return (CTL_RETVAL_COMPLETE);
5150	}
5151
5152	if (length > 0)
5153		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5154
5155	mtx_lock(&ctl_softc->ctl_lock);
5156
5157	/*
5158	 * According to SPC, it is not an error for an intiator to attempt
5159	 * to release a reservation on a LUN that isn't reserved, or that
5160	 * is reserved by another initiator.  The reservation can only be
5161	 * released, though, by the initiator who made it or by one of
5162	 * several reset type events.
5163	 */
5164	if (lun->flags & CTL_LUN_RESERVED) {
5165		if ((ctsio->io_hdr.nexus.initid.id == lun->rsv_nexus.initid.id)
5166		 && (ctsio->io_hdr.nexus.targ_port == lun->rsv_nexus.targ_port)
5167		 && (ctsio->io_hdr.nexus.targ_target.id ==
5168		     lun->rsv_nexus.targ_target.id)) {
5169			lun->flags &= ~CTL_LUN_RESERVED;
5170		}
5171	}
5172
5173	ctsio->scsi_status = SCSI_STATUS_OK;
5174	ctsio->io_hdr.status = CTL_SUCCESS;
5175
5176	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5177		free(ctsio->kern_data_ptr, M_CTL);
5178		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5179	}
5180
5181	mtx_unlock(&ctl_softc->ctl_lock);
5182
5183	ctl_done((union ctl_io *)ctsio);
5184	return (CTL_RETVAL_COMPLETE);
5185}
5186
5187int
5188ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5189{
5190	int extent, thirdparty, longid;
5191	int resv_id, length;
5192	uint64_t thirdparty_id;
5193	struct ctl_softc *ctl_softc;
5194	struct ctl_lun *lun;
5195
5196	extent = 0;
5197	thirdparty = 0;
5198	longid = 0;
5199	resv_id = 0;
5200	length = 0;
5201	thirdparty_id = 0;
5202
5203	CTL_DEBUG_PRINT(("ctl_reserve\n"));
5204
5205	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5206	ctl_softc = control_softc;
5207
5208	switch (ctsio->cdb[0]) {
5209	case RESERVE: {
5210		struct scsi_reserve *cdb;
5211
5212		cdb = (struct scsi_reserve *)ctsio->cdb;
5213		if ((cdb->byte2 & 0x1f) != 0) {
5214			ctl_set_invalid_field(ctsio,
5215					      /*sks_valid*/ 1,
5216					      /*command*/ 1,
5217					      /*field*/ 1,
5218					      /*bit_valid*/ 0,
5219					      /*bit*/ 0);
5220			ctl_done((union ctl_io *)ctsio);
5221			return (CTL_RETVAL_COMPLETE);
5222		}
5223		resv_id = cdb->resv_id;
5224		length = scsi_2btoul(cdb->length);
5225		break;
5226	}
5227	case RESERVE_10: {
5228		struct scsi_reserve_10 *cdb;
5229
5230		cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5231
5232		if ((cdb->byte2 & SR10_EXTENT) != 0) {
5233			ctl_set_invalid_field(ctsio,
5234					      /*sks_valid*/ 1,
5235					      /*command*/ 1,
5236					      /*field*/ 1,
5237					      /*bit_valid*/ 1,
5238					      /*bit*/ 0);
5239			ctl_done((union ctl_io *)ctsio);
5240			return (CTL_RETVAL_COMPLETE);
5241		}
5242		if ((cdb->byte2 & SR10_3RDPTY) != 0) {
5243			ctl_set_invalid_field(ctsio,
5244					      /*sks_valid*/ 1,
5245					      /*command*/ 1,
5246					      /*field*/ 1,
5247					      /*bit_valid*/ 1,
5248					      /*bit*/ 4);
5249			ctl_done((union ctl_io *)ctsio);
5250			return (CTL_RETVAL_COMPLETE);
5251		}
5252		if (cdb->byte2 & SR10_LONGID)
5253			longid = 1;
5254		else
5255			thirdparty_id = cdb->thirdparty_id;
5256
5257		resv_id = cdb->resv_id;
5258		length = scsi_2btoul(cdb->length);
5259		break;
5260	}
5261	}
5262
5263	/*
5264	 * XXX KDM right now, we only support LUN reservation.  We don't
5265	 * support 3rd party reservations, or extent reservations, which
5266	 * might actually need the parameter list.  If we've gotten this
5267	 * far, we've got a LUN reservation.  Anything else got kicked out
5268	 * above.  So, according to SPC, ignore the length.
5269	 */
5270	length = 0;
5271
5272	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5273	 && (length > 0)) {
5274		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5275		ctsio->kern_data_len = length;
5276		ctsio->kern_total_len = length;
5277		ctsio->kern_data_resid = 0;
5278		ctsio->kern_rel_offset = 0;
5279		ctsio->kern_sg_entries = 0;
5280		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5281		ctsio->be_move_done = ctl_config_move_done;
5282		ctl_datamove((union ctl_io *)ctsio);
5283
5284		return (CTL_RETVAL_COMPLETE);
5285	}
5286
5287	if (length > 0)
5288		thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5289
5290	mtx_lock(&ctl_softc->ctl_lock);
5291	if (lun->flags & CTL_LUN_RESERVED) {
5292		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
5293		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
5294		 || (ctsio->io_hdr.nexus.targ_target.id !=
5295		     lun->rsv_nexus.targ_target.id)) {
5296			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
5297			ctsio->io_hdr.status = CTL_SCSI_ERROR;
5298			goto bailout;
5299		}
5300	}
5301
5302	lun->flags |= CTL_LUN_RESERVED;
5303	lun->rsv_nexus = ctsio->io_hdr.nexus;
5304
5305	ctsio->scsi_status = SCSI_STATUS_OK;
5306	ctsio->io_hdr.status = CTL_SUCCESS;
5307
5308bailout:
5309	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5310		free(ctsio->kern_data_ptr, M_CTL);
5311		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5312	}
5313
5314	mtx_unlock(&ctl_softc->ctl_lock);
5315
5316	ctl_done((union ctl_io *)ctsio);
5317	return (CTL_RETVAL_COMPLETE);
5318}
5319
5320int
5321ctl_start_stop(struct ctl_scsiio *ctsio)
5322{
5323	struct scsi_start_stop_unit *cdb;
5324	struct ctl_lun *lun;
5325	struct ctl_softc *ctl_softc;
5326	int retval;
5327
5328	CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5329
5330	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5331	ctl_softc = control_softc;
5332	retval = 0;
5333
5334	cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5335
5336	/*
5337	 * XXX KDM
5338	 * We don't support the immediate bit on a stop unit.  In order to
5339	 * do that, we would need to code up a way to know that a stop is
5340	 * pending, and hold off any new commands until it completes, one
5341	 * way or another.  Then we could accept or reject those commands
5342	 * depending on its status.  We would almost need to do the reverse
5343	 * of what we do below for an immediate start -- return the copy of
5344	 * the ctl_io to the FETD with status to send to the host (and to
5345	 * free the copy!) and then free the original I/O once the stop
5346	 * actually completes.  That way, the OOA queue mechanism can work
5347	 * to block commands that shouldn't proceed.  Another alternative
5348	 * would be to put the copy in the queue in place of the original,
5349	 * and return the original back to the caller.  That could be
5350	 * slightly safer..
5351	 */
5352	if ((cdb->byte2 & SSS_IMMED)
5353	 && ((cdb->how & SSS_START) == 0)) {
5354		ctl_set_invalid_field(ctsio,
5355				      /*sks_valid*/ 1,
5356				      /*command*/ 1,
5357				      /*field*/ 1,
5358				      /*bit_valid*/ 1,
5359				      /*bit*/ 0);
5360		ctl_done((union ctl_io *)ctsio);
5361		return (CTL_RETVAL_COMPLETE);
5362	}
5363
5364	/*
5365	 * We don't support the power conditions field.  We need to check
5366	 * this prior to checking the load/eject and start/stop bits.
5367	 */
5368	if ((cdb->how & SSS_PC_MASK) != SSS_PC_START_VALID) {
5369		ctl_set_invalid_field(ctsio,
5370				      /*sks_valid*/ 1,
5371				      /*command*/ 1,
5372				      /*field*/ 4,
5373				      /*bit_valid*/ 1,
5374				      /*bit*/ 4);
5375		ctl_done((union ctl_io *)ctsio);
5376		return (CTL_RETVAL_COMPLETE);
5377	}
5378
5379	/*
5380	 * Media isn't removable, so we can't load or eject it.
5381	 */
5382	if ((cdb->how & SSS_LOEJ) != 0) {
5383		ctl_set_invalid_field(ctsio,
5384				      /*sks_valid*/ 1,
5385				      /*command*/ 1,
5386				      /*field*/ 4,
5387				      /*bit_valid*/ 1,
5388				      /*bit*/ 1);
5389		ctl_done((union ctl_io *)ctsio);
5390		return (CTL_RETVAL_COMPLETE);
5391	}
5392
5393	if ((lun->flags & CTL_LUN_PR_RESERVED)
5394	 && ((cdb->how & SSS_START)==0)) {
5395		uint32_t residx;
5396
5397		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
5398		if (!lun->per_res[residx].registered
5399		 || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5400
5401			ctl_set_reservation_conflict(ctsio);
5402			ctl_done((union ctl_io *)ctsio);
5403			return (CTL_RETVAL_COMPLETE);
5404		}
5405	}
5406
5407	/*
5408	 * If there is no backend on this device, we can't start or stop
5409	 * it.  In theory we shouldn't get any start/stop commands in the
5410	 * first place at this level if the LUN doesn't have a backend.
5411	 * That should get stopped by the command decode code.
5412	 */
5413	if (lun->backend == NULL) {
5414		ctl_set_invalid_opcode(ctsio);
5415		ctl_done((union ctl_io *)ctsio);
5416		return (CTL_RETVAL_COMPLETE);
5417	}
5418
5419	/*
5420	 * XXX KDM Copan-specific offline behavior.
5421	 * Figure out a reasonable way to port this?
5422	 */
5423#ifdef NEEDTOPORT
5424	mtx_lock(&ctl_softc->ctl_lock);
5425
5426	if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5427	 && (lun->flags & CTL_LUN_OFFLINE)) {
5428		/*
5429		 * If the LUN is offline, and the on/offline bit isn't set,
5430		 * reject the start or stop.  Otherwise, let it through.
5431		 */
5432		mtx_unlock(&ctl_softc->ctl_lock);
5433		ctl_set_lun_not_ready(ctsio);
5434		ctl_done((union ctl_io *)ctsio);
5435	} else {
5436		mtx_unlock(&ctl_softc->ctl_lock);
5437#endif /* NEEDTOPORT */
5438		/*
5439		 * This could be a start or a stop when we're online,
5440		 * or a stop/offline or start/online.  A start or stop when
5441		 * we're offline is covered in the case above.
5442		 */
5443		/*
5444		 * In the non-immediate case, we send the request to
5445		 * the backend and return status to the user when
5446		 * it is done.
5447		 *
5448		 * In the immediate case, we allocate a new ctl_io
5449		 * to hold a copy of the request, and send that to
5450		 * the backend.  We then set good status on the
5451		 * user's request and return it immediately.
5452		 */
5453		if (cdb->byte2 & SSS_IMMED) {
5454			union ctl_io *new_io;
5455
5456			new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5457			if (new_io == NULL) {
5458				ctl_set_busy(ctsio);
5459				ctl_done((union ctl_io *)ctsio);
5460			} else {
5461				ctl_copy_io((union ctl_io *)ctsio,
5462					    new_io);
5463				retval = lun->backend->config_write(new_io);
5464				ctl_set_success(ctsio);
5465				ctl_done((union ctl_io *)ctsio);
5466			}
5467		} else {
5468			retval = lun->backend->config_write(
5469				(union ctl_io *)ctsio);
5470		}
5471#ifdef NEEDTOPORT
5472	}
5473#endif
5474	return (retval);
5475}
5476
5477/*
5478 * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5479 * we don't really do anything with the LBA and length fields if the user
5480 * passes them in.  Instead we'll just flush out the cache for the entire
5481 * LUN.
5482 */
5483int
5484ctl_sync_cache(struct ctl_scsiio *ctsio)
5485{
5486	struct ctl_lun *lun;
5487	struct ctl_softc *ctl_softc;
5488	uint64_t starting_lba;
5489	uint32_t block_count;
5490	int reladr, immed;
5491	int retval;
5492
5493	CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5494
5495	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5496	ctl_softc = control_softc;
5497	retval = 0;
5498	reladr = 0;
5499	immed = 0;
5500
5501	switch (ctsio->cdb[0]) {
5502	case SYNCHRONIZE_CACHE: {
5503		struct scsi_sync_cache *cdb;
5504		cdb = (struct scsi_sync_cache *)ctsio->cdb;
5505
5506		if (cdb->byte2 & SSC_RELADR)
5507			reladr = 1;
5508
5509		if (cdb->byte2 & SSC_IMMED)
5510			immed = 1;
5511
5512		starting_lba = scsi_4btoul(cdb->begin_lba);
5513		block_count = scsi_2btoul(cdb->lb_count);
5514		break;
5515	}
5516	case SYNCHRONIZE_CACHE_16: {
5517		struct scsi_sync_cache_16 *cdb;
5518		cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5519
5520		if (cdb->byte2 & SSC_RELADR)
5521			reladr = 1;
5522
5523		if (cdb->byte2 & SSC_IMMED)
5524			immed = 1;
5525
5526		starting_lba = scsi_8btou64(cdb->begin_lba);
5527		block_count = scsi_4btoul(cdb->lb_count);
5528		break;
5529	}
5530	default:
5531		ctl_set_invalid_opcode(ctsio);
5532		ctl_done((union ctl_io *)ctsio);
5533		goto bailout;
5534		break; /* NOTREACHED */
5535	}
5536
5537	if (immed) {
5538		/*
5539		 * We don't support the immediate bit.  Since it's in the
5540		 * same place for the 10 and 16 byte SYNCHRONIZE CACHE
5541		 * commands, we can just return the same error in either
5542		 * case.
5543		 */
5544		ctl_set_invalid_field(ctsio,
5545				      /*sks_valid*/ 1,
5546				      /*command*/ 1,
5547				      /*field*/ 1,
5548				      /*bit_valid*/ 1,
5549				      /*bit*/ 1);
5550		ctl_done((union ctl_io *)ctsio);
5551		goto bailout;
5552	}
5553
5554	if (reladr) {
5555		/*
5556		 * We don't support the reladr bit either.  It can only be
5557		 * used with linked commands, and we don't support linked
5558		 * commands.  Since the bit is in the same place for the
5559		 * 10 and 16 byte SYNCHRONIZE CACHE * commands, we can
5560		 * just return the same error in either case.
5561		 */
5562		ctl_set_invalid_field(ctsio,
5563				      /*sks_valid*/ 1,
5564				      /*command*/ 1,
5565				      /*field*/ 1,
5566				      /*bit_valid*/ 1,
5567				      /*bit*/ 0);
5568		ctl_done((union ctl_io *)ctsio);
5569		goto bailout;
5570	}
5571
5572	/*
5573	 * We check the LBA and length, but don't do anything with them.
5574	 * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5575	 * get flushed.  This check will just help satisfy anyone who wants
5576	 * to see an error for an out of range LBA.
5577	 */
5578	if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5579		ctl_set_lba_out_of_range(ctsio);
5580		ctl_done((union ctl_io *)ctsio);
5581		goto bailout;
5582	}
5583
5584	/*
5585	 * If this LUN has no backend, we can't flush the cache anyway.
5586	 */
5587	if (lun->backend == NULL) {
5588		ctl_set_invalid_opcode(ctsio);
5589		ctl_done((union ctl_io *)ctsio);
5590		goto bailout;
5591	}
5592
5593	/*
5594	 * Check to see whether we're configured to send the SYNCHRONIZE
5595	 * CACHE command directly to the back end.
5596	 */
5597	mtx_lock(&ctl_softc->ctl_lock);
5598	if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC)
5599	 && (++(lun->sync_count) >= lun->sync_interval)) {
5600		lun->sync_count = 0;
5601		mtx_unlock(&ctl_softc->ctl_lock);
5602		retval = lun->backend->config_write((union ctl_io *)ctsio);
5603	} else {
5604		mtx_unlock(&ctl_softc->ctl_lock);
5605		ctl_set_success(ctsio);
5606		ctl_done((union ctl_io *)ctsio);
5607	}
5608
5609bailout:
5610
5611	return (retval);
5612}
5613
5614int
5615ctl_format(struct ctl_scsiio *ctsio)
5616{
5617	struct scsi_format *cdb;
5618	struct ctl_lun *lun;
5619	struct ctl_softc *ctl_softc;
5620	int length, defect_list_len;
5621
5622	CTL_DEBUG_PRINT(("ctl_format\n"));
5623
5624	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5625	ctl_softc = control_softc;
5626
5627	cdb = (struct scsi_format *)ctsio->cdb;
5628
5629	length = 0;
5630	if (cdb->byte2 & SF_FMTDATA) {
5631		if (cdb->byte2 & SF_LONGLIST)
5632			length = sizeof(struct scsi_format_header_long);
5633		else
5634			length = sizeof(struct scsi_format_header_short);
5635	}
5636
5637	if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5638	 && (length > 0)) {
5639		ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5640		ctsio->kern_data_len = length;
5641		ctsio->kern_total_len = length;
5642		ctsio->kern_data_resid = 0;
5643		ctsio->kern_rel_offset = 0;
5644		ctsio->kern_sg_entries = 0;
5645		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5646		ctsio->be_move_done = ctl_config_move_done;
5647		ctl_datamove((union ctl_io *)ctsio);
5648
5649		return (CTL_RETVAL_COMPLETE);
5650	}
5651
5652	defect_list_len = 0;
5653
5654	if (cdb->byte2 & SF_FMTDATA) {
5655		if (cdb->byte2 & SF_LONGLIST) {
5656			struct scsi_format_header_long *header;
5657
5658			header = (struct scsi_format_header_long *)
5659				ctsio->kern_data_ptr;
5660
5661			defect_list_len = scsi_4btoul(header->defect_list_len);
5662			if (defect_list_len != 0) {
5663				ctl_set_invalid_field(ctsio,
5664						      /*sks_valid*/ 1,
5665						      /*command*/ 0,
5666						      /*field*/ 2,
5667						      /*bit_valid*/ 0,
5668						      /*bit*/ 0);
5669				goto bailout;
5670			}
5671		} else {
5672			struct scsi_format_header_short *header;
5673
5674			header = (struct scsi_format_header_short *)
5675				ctsio->kern_data_ptr;
5676
5677			defect_list_len = scsi_2btoul(header->defect_list_len);
5678			if (defect_list_len != 0) {
5679				ctl_set_invalid_field(ctsio,
5680						      /*sks_valid*/ 1,
5681						      /*command*/ 0,
5682						      /*field*/ 2,
5683						      /*bit_valid*/ 0,
5684						      /*bit*/ 0);
5685				goto bailout;
5686			}
5687		}
5688	}
5689
5690	/*
5691	 * The format command will clear out the "Medium format corrupted"
5692	 * status if set by the configuration code.  That status is really
5693	 * just a way to notify the host that we have lost the media, and
5694	 * get them to issue a command that will basically make them think
5695	 * they're blowing away the media.
5696	 */
5697	mtx_lock(&ctl_softc->ctl_lock);
5698	lun->flags &= ~CTL_LUN_INOPERABLE;
5699	mtx_unlock(&ctl_softc->ctl_lock);
5700
5701	ctsio->scsi_status = SCSI_STATUS_OK;
5702	ctsio->io_hdr.status = CTL_SUCCESS;
5703bailout:
5704
5705	if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5706		free(ctsio->kern_data_ptr, M_CTL);
5707		ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5708	}
5709
5710	ctl_done((union ctl_io *)ctsio);
5711	return (CTL_RETVAL_COMPLETE);
5712}
5713
5714int
5715ctl_write_buffer(struct ctl_scsiio *ctsio)
5716{
5717	struct scsi_write_buffer *cdb;
5718	struct copan_page_header *header;
5719	struct ctl_lun *lun;
5720	struct ctl_softc *ctl_softc;
5721	int buffer_offset, len;
5722	int retval;
5723
5724	header = NULL;
5725
5726	retval = CTL_RETVAL_COMPLETE;
5727
5728	CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5729
5730	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5731	ctl_softc = control_softc;
5732	cdb = (struct scsi_write_buffer *)ctsio->cdb;
5733
5734	if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5735		ctl_set_invalid_field(ctsio,
5736				      /*sks_valid*/ 1,
5737				      /*command*/ 1,
5738				      /*field*/ 1,
5739				      /*bit_valid*/ 1,
5740				      /*bit*/ 4);
5741		ctl_done((union ctl_io *)ctsio);
5742		return (CTL_RETVAL_COMPLETE);
5743	}
5744	if (cdb->buffer_id != 0) {
5745		ctl_set_invalid_field(ctsio,
5746				      /*sks_valid*/ 1,
5747				      /*command*/ 1,
5748				      /*field*/ 2,
5749				      /*bit_valid*/ 0,
5750				      /*bit*/ 0);
5751		ctl_done((union ctl_io *)ctsio);
5752		return (CTL_RETVAL_COMPLETE);
5753	}
5754
5755	len = scsi_3btoul(cdb->length);
5756	buffer_offset = scsi_3btoul(cdb->offset);
5757
5758	if (len > sizeof(lun->write_buffer)) {
5759		ctl_set_invalid_field(ctsio,
5760				      /*sks_valid*/ 1,
5761				      /*command*/ 1,
5762				      /*field*/ 6,
5763				      /*bit_valid*/ 0,
5764				      /*bit*/ 0);
5765		ctl_done((union ctl_io *)ctsio);
5766		return (CTL_RETVAL_COMPLETE);
5767	}
5768
5769	if (buffer_offset != 0) {
5770		ctl_set_invalid_field(ctsio,
5771				      /*sks_valid*/ 1,
5772				      /*command*/ 1,
5773				      /*field*/ 3,
5774				      /*bit_valid*/ 0,
5775				      /*bit*/ 0);
5776		ctl_done((union ctl_io *)ctsio);
5777		return (CTL_RETVAL_COMPLETE);
5778	}
5779
5780	/*
5781	 * If we've got a kernel request that hasn't been malloced yet,
5782	 * malloc it and tell the caller the data buffer is here.
5783	 */
5784	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5785		ctsio->kern_data_ptr = lun->write_buffer;
5786		ctsio->kern_data_len = len;
5787		ctsio->kern_total_len = len;
5788		ctsio->kern_data_resid = 0;
5789		ctsio->kern_rel_offset = 0;
5790		ctsio->kern_sg_entries = 0;
5791		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5792		ctsio->be_move_done = ctl_config_move_done;
5793		ctl_datamove((union ctl_io *)ctsio);
5794
5795		return (CTL_RETVAL_COMPLETE);
5796	}
5797
5798	ctl_done((union ctl_io *)ctsio);
5799
5800	return (CTL_RETVAL_COMPLETE);
5801}
5802
5803int
5804ctl_write_same(struct ctl_scsiio *ctsio)
5805{
5806	struct ctl_lun *lun;
5807	struct ctl_lba_len_flags *lbalen;
5808	uint64_t lba;
5809	uint32_t num_blocks;
5810	int len, retval;
5811	uint8_t byte2;
5812
5813	retval = CTL_RETVAL_COMPLETE;
5814
5815	CTL_DEBUG_PRINT(("ctl_write_same\n"));
5816
5817	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5818
5819	switch (ctsio->cdb[0]) {
5820	case WRITE_SAME_10: {
5821		struct scsi_write_same_10 *cdb;
5822
5823		cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5824
5825		lba = scsi_4btoul(cdb->addr);
5826		num_blocks = scsi_2btoul(cdb->length);
5827		byte2 = cdb->byte2;
5828		break;
5829	}
5830	case WRITE_SAME_16: {
5831		struct scsi_write_same_16 *cdb;
5832
5833		cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5834
5835		lba = scsi_8btou64(cdb->addr);
5836		num_blocks = scsi_4btoul(cdb->length);
5837		byte2 = cdb->byte2;
5838		break;
5839	}
5840	default:
5841		/*
5842		 * We got a command we don't support.  This shouldn't
5843		 * happen, commands should be filtered out above us.
5844		 */
5845		ctl_set_invalid_opcode(ctsio);
5846		ctl_done((union ctl_io *)ctsio);
5847
5848		return (CTL_RETVAL_COMPLETE);
5849		break; /* NOTREACHED */
5850	}
5851
5852	/*
5853	 * The first check is to make sure we're in bounds, the second
5854	 * check is to catch wrap-around problems.  If the lba + num blocks
5855	 * is less than the lba, then we've wrapped around and the block
5856	 * range is invalid anyway.
5857	 */
5858	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5859	 || ((lba + num_blocks) < lba)) {
5860		ctl_set_lba_out_of_range(ctsio);
5861		ctl_done((union ctl_io *)ctsio);
5862		return (CTL_RETVAL_COMPLETE);
5863	}
5864
5865	/* Zero number of blocks means "to the last logical block" */
5866	if (num_blocks == 0) {
5867		if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5868			ctl_set_invalid_field(ctsio,
5869					      /*sks_valid*/ 0,
5870					      /*command*/ 1,
5871					      /*field*/ 0,
5872					      /*bit_valid*/ 0,
5873					      /*bit*/ 0);
5874			ctl_done((union ctl_io *)ctsio);
5875			return (CTL_RETVAL_COMPLETE);
5876		}
5877		num_blocks = (lun->be_lun->maxlba + 1) - lba;
5878	}
5879
5880	len = lun->be_lun->blocksize;
5881
5882	/*
5883	 * If we've got a kernel request that hasn't been malloced yet,
5884	 * malloc it and tell the caller the data buffer is here.
5885	 */
5886	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5887		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5888		ctsio->kern_data_len = len;
5889		ctsio->kern_total_len = len;
5890		ctsio->kern_data_resid = 0;
5891		ctsio->kern_rel_offset = 0;
5892		ctsio->kern_sg_entries = 0;
5893		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5894		ctsio->be_move_done = ctl_config_move_done;
5895		ctl_datamove((union ctl_io *)ctsio);
5896
5897		return (CTL_RETVAL_COMPLETE);
5898	}
5899
5900	lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5901	lbalen->lba = lba;
5902	lbalen->len = num_blocks;
5903	lbalen->flags = byte2;
5904	retval = lun->backend->config_write((union ctl_io *)ctsio);
5905
5906	return (retval);
5907}
5908
5909int
5910ctl_unmap(struct ctl_scsiio *ctsio)
5911{
5912	struct ctl_lun *lun;
5913	struct scsi_unmap *cdb;
5914	struct ctl_ptr_len_flags *ptrlen;
5915	struct scsi_unmap_header *hdr;
5916	struct scsi_unmap_desc *buf, *end;
5917	uint64_t lba;
5918	uint32_t num_blocks;
5919	int len, retval;
5920	uint8_t byte2;
5921
5922	retval = CTL_RETVAL_COMPLETE;
5923
5924	CTL_DEBUG_PRINT(("ctl_unmap\n"));
5925
5926	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5927	cdb = (struct scsi_unmap *)ctsio->cdb;
5928
5929	len = scsi_2btoul(cdb->length);
5930	byte2 = cdb->byte2;
5931
5932	/*
5933	 * If we've got a kernel request that hasn't been malloced yet,
5934	 * malloc it and tell the caller the data buffer is here.
5935	 */
5936	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5937		ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5938		ctsio->kern_data_len = len;
5939		ctsio->kern_total_len = len;
5940		ctsio->kern_data_resid = 0;
5941		ctsio->kern_rel_offset = 0;
5942		ctsio->kern_sg_entries = 0;
5943		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5944		ctsio->be_move_done = ctl_config_move_done;
5945		ctl_datamove((union ctl_io *)ctsio);
5946
5947		return (CTL_RETVAL_COMPLETE);
5948	}
5949
5950	len = ctsio->kern_total_len - ctsio->kern_data_resid;
5951	hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5952	if (len < sizeof (*hdr) ||
5953	    len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5954	    len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5955	    scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5956		ctl_set_invalid_field(ctsio,
5957				      /*sks_valid*/ 0,
5958				      /*command*/ 0,
5959				      /*field*/ 0,
5960				      /*bit_valid*/ 0,
5961				      /*bit*/ 0);
5962		ctl_done((union ctl_io *)ctsio);
5963		return (CTL_RETVAL_COMPLETE);
5964	}
5965	len = scsi_2btoul(hdr->desc_length);
5966	buf = (struct scsi_unmap_desc *)(hdr + 1);
5967	end = buf + len / sizeof(*buf);
5968
5969	ptrlen = (struct ctl_ptr_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5970	ptrlen->ptr = (void *)buf;
5971	ptrlen->len = len;
5972	ptrlen->flags = byte2;
5973
5974	for (; buf < end; buf++) {
5975		lba = scsi_8btou64(buf->lba);
5976		num_blocks = scsi_4btoul(buf->length);
5977		if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5978		 || ((lba + num_blocks) < lba)) {
5979			ctl_set_lba_out_of_range(ctsio);
5980			ctl_done((union ctl_io *)ctsio);
5981			return (CTL_RETVAL_COMPLETE);
5982		}
5983	}
5984
5985	retval = lun->backend->config_write((union ctl_io *)ctsio);
5986
5987	return (retval);
5988}
5989
5990/*
5991 * Note that this function currently doesn't actually do anything inside
5992 * CTL to enforce things if the DQue bit is turned on.
5993 *
5994 * Also note that this function can't be used in the default case, because
5995 * the DQue bit isn't set in the changeable mask for the control mode page
5996 * anyway.  This is just here as an example for how to implement a page
5997 * handler, and a placeholder in case we want to allow the user to turn
5998 * tagged queueing on and off.
5999 *
6000 * The D_SENSE bit handling is functional, however, and will turn
6001 * descriptor sense on and off for a given LUN.
6002 */
6003int
6004ctl_control_page_handler(struct ctl_scsiio *ctsio,
6005			 struct ctl_page_index *page_index, uint8_t *page_ptr)
6006{
6007	struct scsi_control_page *current_cp, *saved_cp, *user_cp;
6008	struct ctl_lun *lun;
6009	struct ctl_softc *softc;
6010	int set_ua;
6011	uint32_t initidx;
6012
6013	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6014	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6015	set_ua = 0;
6016
6017	user_cp = (struct scsi_control_page *)page_ptr;
6018	current_cp = (struct scsi_control_page *)
6019		(page_index->page_data + (page_index->page_len *
6020		CTL_PAGE_CURRENT));
6021	saved_cp = (struct scsi_control_page *)
6022		(page_index->page_data + (page_index->page_len *
6023		CTL_PAGE_SAVED));
6024
6025	softc = control_softc;
6026
6027	mtx_lock(&softc->ctl_lock);
6028	if (((current_cp->rlec & SCP_DSENSE) == 0)
6029	 && ((user_cp->rlec & SCP_DSENSE) != 0)) {
6030		/*
6031		 * Descriptor sense is currently turned off and the user
6032		 * wants to turn it on.
6033		 */
6034		current_cp->rlec |= SCP_DSENSE;
6035		saved_cp->rlec |= SCP_DSENSE;
6036		lun->flags |= CTL_LUN_SENSE_DESC;
6037		set_ua = 1;
6038	} else if (((current_cp->rlec & SCP_DSENSE) != 0)
6039		&& ((user_cp->rlec & SCP_DSENSE) == 0)) {
6040		/*
6041		 * Descriptor sense is currently turned on, and the user
6042		 * wants to turn it off.
6043		 */
6044		current_cp->rlec &= ~SCP_DSENSE;
6045		saved_cp->rlec &= ~SCP_DSENSE;
6046		lun->flags &= ~CTL_LUN_SENSE_DESC;
6047		set_ua = 1;
6048	}
6049	if (current_cp->queue_flags & SCP_QUEUE_DQUE) {
6050		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6051#ifdef NEEDTOPORT
6052			csevent_log(CSC_CTL | CSC_SHELF_SW |
6053				    CTL_UNTAG_TO_UNTAG,
6054				    csevent_LogType_Trace,
6055				    csevent_Severity_Information,
6056				    csevent_AlertLevel_Green,
6057				    csevent_FRU_Firmware,
6058				    csevent_FRU_Unknown,
6059				    "Received untagged to untagged transition");
6060#endif /* NEEDTOPORT */
6061		} else {
6062#ifdef NEEDTOPORT
6063			csevent_log(CSC_CTL | CSC_SHELF_SW |
6064				    CTL_UNTAG_TO_TAG,
6065				    csevent_LogType_ConfigChange,
6066				    csevent_Severity_Information,
6067				    csevent_AlertLevel_Green,
6068				    csevent_FRU_Firmware,
6069				    csevent_FRU_Unknown,
6070				    "Received untagged to tagged "
6071				    "queueing transition");
6072#endif /* NEEDTOPORT */
6073
6074			current_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6075			saved_cp->queue_flags &= ~SCP_QUEUE_DQUE;
6076			set_ua = 1;
6077		}
6078	} else {
6079		if (user_cp->queue_flags & SCP_QUEUE_DQUE) {
6080#ifdef NEEDTOPORT
6081			csevent_log(CSC_CTL | CSC_SHELF_SW |
6082				    CTL_TAG_TO_UNTAG,
6083				    csevent_LogType_ConfigChange,
6084				    csevent_Severity_Warning,
6085				    csevent_AlertLevel_Yellow,
6086				    csevent_FRU_Firmware,
6087				    csevent_FRU_Unknown,
6088				    "Received tagged queueing to untagged "
6089				    "transition");
6090#endif /* NEEDTOPORT */
6091
6092			current_cp->queue_flags |= SCP_QUEUE_DQUE;
6093			saved_cp->queue_flags |= SCP_QUEUE_DQUE;
6094			set_ua = 1;
6095		} else {
6096#ifdef NEEDTOPORT
6097			csevent_log(CSC_CTL | CSC_SHELF_SW |
6098				    CTL_TAG_TO_TAG,
6099				    csevent_LogType_Trace,
6100				    csevent_Severity_Information,
6101				    csevent_AlertLevel_Green,
6102				    csevent_FRU_Firmware,
6103				    csevent_FRU_Unknown,
6104				    "Received tagged queueing to tagged "
6105				    "queueing transition");
6106#endif /* NEEDTOPORT */
6107		}
6108	}
6109	if (set_ua != 0) {
6110		int i;
6111		/*
6112		 * Let other initiators know that the mode
6113		 * parameters for this LUN have changed.
6114		 */
6115		for (i = 0; i < CTL_MAX_INITIATORS; i++) {
6116			if (i == initidx)
6117				continue;
6118
6119			lun->pending_sense[i].ua_pending |=
6120				CTL_UA_MODE_CHANGE;
6121		}
6122	}
6123	mtx_unlock(&softc->ctl_lock);
6124
6125	return (0);
6126}
6127
6128int
6129ctl_power_sp_handler(struct ctl_scsiio *ctsio,
6130		     struct ctl_page_index *page_index, uint8_t *page_ptr)
6131{
6132	return (0);
6133}
6134
6135int
6136ctl_power_sp_sense_handler(struct ctl_scsiio *ctsio,
6137			   struct ctl_page_index *page_index, int pc)
6138{
6139	struct copan_power_subpage *page;
6140
6141	page = (struct copan_power_subpage *)page_index->page_data +
6142		(page_index->page_len * pc);
6143
6144	switch (pc) {
6145	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6146		/*
6147		 * We don't update the changable bits for this page.
6148		 */
6149		break;
6150	case SMS_PAGE_CTRL_CURRENT >> 6:
6151	case SMS_PAGE_CTRL_DEFAULT >> 6:
6152	case SMS_PAGE_CTRL_SAVED >> 6:
6153#ifdef NEEDTOPORT
6154		ctl_update_power_subpage(page);
6155#endif
6156		break;
6157	default:
6158#ifdef NEEDTOPORT
6159		EPRINT(0, "Invalid PC %d!!", pc);
6160#endif
6161		break;
6162	}
6163	return (0);
6164}
6165
6166
6167int
6168ctl_aps_sp_handler(struct ctl_scsiio *ctsio,
6169		   struct ctl_page_index *page_index, uint8_t *page_ptr)
6170{
6171	struct copan_aps_subpage *user_sp;
6172	struct copan_aps_subpage *current_sp;
6173	union ctl_modepage_info *modepage_info;
6174	struct ctl_softc *softc;
6175	struct ctl_lun *lun;
6176	int retval;
6177
6178	retval = CTL_RETVAL_COMPLETE;
6179	current_sp = (struct copan_aps_subpage *)(page_index->page_data +
6180		     (page_index->page_len * CTL_PAGE_CURRENT));
6181	softc = control_softc;
6182	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6183
6184	user_sp = (struct copan_aps_subpage *)page_ptr;
6185
6186	modepage_info = (union ctl_modepage_info *)
6187		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6188
6189	modepage_info->header.page_code = page_index->page_code & SMPH_PC_MASK;
6190	modepage_info->header.subpage = page_index->subpage;
6191	modepage_info->aps.lock_active = user_sp->lock_active;
6192
6193	mtx_lock(&softc->ctl_lock);
6194
6195	/*
6196	 * If there is a request to lock the LUN and another LUN is locked
6197	 * this is an error. If the requested LUN is already locked ignore
6198	 * the request. If no LUN is locked attempt to lock it.
6199	 * if there is a request to unlock the LUN and the LUN is currently
6200	 * locked attempt to unlock it. Otherwise ignore the request. i.e.
6201	 * if another LUN is locked or no LUN is locked.
6202	 */
6203	if (user_sp->lock_active & APS_LOCK_ACTIVE) {
6204		if (softc->aps_locked_lun == lun->lun) {
6205			/*
6206			 * This LUN is already locked, so we're done.
6207			 */
6208			retval = CTL_RETVAL_COMPLETE;
6209		} else if (softc->aps_locked_lun == 0) {
6210			/*
6211			 * No one has the lock, pass the request to the
6212			 * backend.
6213			 */
6214			retval = lun->backend->config_write(
6215				(union ctl_io *)ctsio);
6216		} else {
6217			/*
6218			 * Someone else has the lock, throw out the request.
6219			 */
6220			ctl_set_already_locked(ctsio);
6221			free(ctsio->kern_data_ptr, M_CTL);
6222			ctl_done((union ctl_io *)ctsio);
6223
6224			/*
6225			 * Set the return value so that ctl_do_mode_select()
6226			 * won't try to complete the command.  We already
6227			 * completed it here.
6228			 */
6229			retval = CTL_RETVAL_ERROR;
6230		}
6231	} else if (softc->aps_locked_lun == lun->lun) {
6232		/*
6233		 * This LUN is locked, so pass the unlock request to the
6234		 * backend.
6235		 */
6236		retval = lun->backend->config_write((union ctl_io *)ctsio);
6237	}
6238	mtx_unlock(&softc->ctl_lock);
6239
6240	return (retval);
6241}
6242
6243int
6244ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
6245				struct ctl_page_index *page_index,
6246				uint8_t *page_ptr)
6247{
6248	uint8_t *c;
6249	int i;
6250
6251	c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
6252	ctl_time_io_secs =
6253		(c[0] << 8) |
6254		(c[1] << 0) |
6255		0;
6256	CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
6257	printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
6258	printf("page data:");
6259	for (i=0; i<8; i++)
6260		printf(" %.2x",page_ptr[i]);
6261	printf("\n");
6262	return (0);
6263}
6264
6265int
6266ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
6267			       struct ctl_page_index *page_index,
6268			       int pc)
6269{
6270	struct copan_debugconf_subpage *page;
6271
6272	page = (struct copan_debugconf_subpage *)page_index->page_data +
6273		(page_index->page_len * pc);
6274
6275	switch (pc) {
6276	case SMS_PAGE_CTRL_CHANGEABLE >> 6:
6277	case SMS_PAGE_CTRL_DEFAULT >> 6:
6278	case SMS_PAGE_CTRL_SAVED >> 6:
6279		/*
6280		 * We don't update the changable or default bits for this page.
6281		 */
6282		break;
6283	case SMS_PAGE_CTRL_CURRENT >> 6:
6284		page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6285		page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6286		break;
6287	default:
6288#ifdef NEEDTOPORT
6289		EPRINT(0, "Invalid PC %d!!", pc);
6290#endif /* NEEDTOPORT */
6291		break;
6292	}
6293	return (0);
6294}
6295
6296
6297static int
6298ctl_do_mode_select(union ctl_io *io)
6299{
6300	struct scsi_mode_page_header *page_header;
6301	struct ctl_page_index *page_index;
6302	struct ctl_scsiio *ctsio;
6303	int control_dev, page_len;
6304	int page_len_offset, page_len_size;
6305	union ctl_modepage_info *modepage_info;
6306	struct ctl_lun *lun;
6307	int *len_left, *len_used;
6308	int retval, i;
6309
6310	ctsio = &io->scsiio;
6311	page_index = NULL;
6312	page_len = 0;
6313	retval = CTL_RETVAL_COMPLETE;
6314
6315	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6316
6317	if (lun->be_lun->lun_type != T_DIRECT)
6318		control_dev = 1;
6319	else
6320		control_dev = 0;
6321
6322	modepage_info = (union ctl_modepage_info *)
6323		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6324	len_left = &modepage_info->header.len_left;
6325	len_used = &modepage_info->header.len_used;
6326
6327do_next_page:
6328
6329	page_header = (struct scsi_mode_page_header *)
6330		(ctsio->kern_data_ptr + *len_used);
6331
6332	if (*len_left == 0) {
6333		free(ctsio->kern_data_ptr, M_CTL);
6334		ctl_set_success(ctsio);
6335		ctl_done((union ctl_io *)ctsio);
6336		return (CTL_RETVAL_COMPLETE);
6337	} else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6338
6339		free(ctsio->kern_data_ptr, M_CTL);
6340		ctl_set_param_len_error(ctsio);
6341		ctl_done((union ctl_io *)ctsio);
6342		return (CTL_RETVAL_COMPLETE);
6343
6344	} else if ((page_header->page_code & SMPH_SPF)
6345		&& (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6346
6347		free(ctsio->kern_data_ptr, M_CTL);
6348		ctl_set_param_len_error(ctsio);
6349		ctl_done((union ctl_io *)ctsio);
6350		return (CTL_RETVAL_COMPLETE);
6351	}
6352
6353
6354	/*
6355	 * XXX KDM should we do something with the block descriptor?
6356	 */
6357	for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6358
6359		if ((control_dev != 0)
6360		 && (lun->mode_pages.index[i].page_flags &
6361		     CTL_PAGE_FLAG_DISK_ONLY))
6362			continue;
6363
6364		if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6365		    (page_header->page_code & SMPH_PC_MASK))
6366			continue;
6367
6368		/*
6369		 * If neither page has a subpage code, then we've got a
6370		 * match.
6371		 */
6372		if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6373		 && ((page_header->page_code & SMPH_SPF) == 0)) {
6374			page_index = &lun->mode_pages.index[i];
6375			page_len = page_header->page_length;
6376			break;
6377		}
6378
6379		/*
6380		 * If both pages have subpages, then the subpage numbers
6381		 * have to match.
6382		 */
6383		if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6384		  && (page_header->page_code & SMPH_SPF)) {
6385			struct scsi_mode_page_header_sp *sph;
6386
6387			sph = (struct scsi_mode_page_header_sp *)page_header;
6388
6389			if (lun->mode_pages.index[i].subpage ==
6390			    sph->subpage) {
6391				page_index = &lun->mode_pages.index[i];
6392				page_len = scsi_2btoul(sph->page_length);
6393				break;
6394			}
6395		}
6396	}
6397
6398	/*
6399	 * If we couldn't find the page, or if we don't have a mode select
6400	 * handler for it, send back an error to the user.
6401	 */
6402	if ((page_index == NULL)
6403	 || (page_index->select_handler == NULL)) {
6404		ctl_set_invalid_field(ctsio,
6405				      /*sks_valid*/ 1,
6406				      /*command*/ 0,
6407				      /*field*/ *len_used,
6408				      /*bit_valid*/ 0,
6409				      /*bit*/ 0);
6410		free(ctsio->kern_data_ptr, M_CTL);
6411		ctl_done((union ctl_io *)ctsio);
6412		return (CTL_RETVAL_COMPLETE);
6413	}
6414
6415	if (page_index->page_code & SMPH_SPF) {
6416		page_len_offset = 2;
6417		page_len_size = 2;
6418	} else {
6419		page_len_size = 1;
6420		page_len_offset = 1;
6421	}
6422
6423	/*
6424	 * If the length the initiator gives us isn't the one we specify in
6425	 * the mode page header, or if they didn't specify enough data in
6426	 * the CDB to avoid truncating this page, kick out the request.
6427	 */
6428	if ((page_len != (page_index->page_len - page_len_offset -
6429			  page_len_size))
6430	 || (*len_left < page_index->page_len)) {
6431
6432
6433		ctl_set_invalid_field(ctsio,
6434				      /*sks_valid*/ 1,
6435				      /*command*/ 0,
6436				      /*field*/ *len_used + page_len_offset,
6437				      /*bit_valid*/ 0,
6438				      /*bit*/ 0);
6439		free(ctsio->kern_data_ptr, M_CTL);
6440		ctl_done((union ctl_io *)ctsio);
6441		return (CTL_RETVAL_COMPLETE);
6442	}
6443
6444	/*
6445	 * Run through the mode page, checking to make sure that the bits
6446	 * the user changed are actually legal for him to change.
6447	 */
6448	for (i = 0; i < page_index->page_len; i++) {
6449		uint8_t *user_byte, *change_mask, *current_byte;
6450		int bad_bit;
6451		int j;
6452
6453		user_byte = (uint8_t *)page_header + i;
6454		change_mask = page_index->page_data +
6455			      (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6456		current_byte = page_index->page_data +
6457			       (page_index->page_len * CTL_PAGE_CURRENT) + i;
6458
6459		/*
6460		 * Check to see whether the user set any bits in this byte
6461		 * that he is not allowed to set.
6462		 */
6463		if ((*user_byte & ~(*change_mask)) ==
6464		    (*current_byte & ~(*change_mask)))
6465			continue;
6466
6467		/*
6468		 * Go through bit by bit to determine which one is illegal.
6469		 */
6470		bad_bit = 0;
6471		for (j = 7; j >= 0; j--) {
6472			if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6473			    (((1 << i) & ~(*change_mask)) & *current_byte)) {
6474				bad_bit = i;
6475				break;
6476			}
6477		}
6478		ctl_set_invalid_field(ctsio,
6479				      /*sks_valid*/ 1,
6480				      /*command*/ 0,
6481				      /*field*/ *len_used + i,
6482				      /*bit_valid*/ 1,
6483				      /*bit*/ bad_bit);
6484		free(ctsio->kern_data_ptr, M_CTL);
6485		ctl_done((union ctl_io *)ctsio);
6486		return (CTL_RETVAL_COMPLETE);
6487	}
6488
6489	/*
6490	 * Decrement these before we call the page handler, since we may
6491	 * end up getting called back one way or another before the handler
6492	 * returns to this context.
6493	 */
6494	*len_left -= page_index->page_len;
6495	*len_used += page_index->page_len;
6496
6497	retval = page_index->select_handler(ctsio, page_index,
6498					    (uint8_t *)page_header);
6499
6500	/*
6501	 * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6502	 * wait until this queued command completes to finish processing
6503	 * the mode page.  If it returns anything other than
6504	 * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6505	 * already set the sense information, freed the data pointer, and
6506	 * completed the io for us.
6507	 */
6508	if (retval != CTL_RETVAL_COMPLETE)
6509		goto bailout_no_done;
6510
6511	/*
6512	 * If the initiator sent us more than one page, parse the next one.
6513	 */
6514	if (*len_left > 0)
6515		goto do_next_page;
6516
6517	ctl_set_success(ctsio);
6518	free(ctsio->kern_data_ptr, M_CTL);
6519	ctl_done((union ctl_io *)ctsio);
6520
6521bailout_no_done:
6522
6523	return (CTL_RETVAL_COMPLETE);
6524
6525}
6526
6527int
6528ctl_mode_select(struct ctl_scsiio *ctsio)
6529{
6530	int param_len, pf, sp;
6531	int header_size, bd_len;
6532	int len_left, len_used;
6533	struct ctl_page_index *page_index;
6534	struct ctl_lun *lun;
6535	int control_dev, page_len;
6536	union ctl_modepage_info *modepage_info;
6537	int retval;
6538
6539	pf = 0;
6540	sp = 0;
6541	page_len = 0;
6542	len_used = 0;
6543	len_left = 0;
6544	retval = 0;
6545	bd_len = 0;
6546	page_index = NULL;
6547
6548	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6549
6550	if (lun->be_lun->lun_type != T_DIRECT)
6551		control_dev = 1;
6552	else
6553		control_dev = 0;
6554
6555	switch (ctsio->cdb[0]) {
6556	case MODE_SELECT_6: {
6557		struct scsi_mode_select_6 *cdb;
6558
6559		cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6560
6561		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6562		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6563
6564		param_len = cdb->length;
6565		header_size = sizeof(struct scsi_mode_header_6);
6566		break;
6567	}
6568	case MODE_SELECT_10: {
6569		struct scsi_mode_select_10 *cdb;
6570
6571		cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6572
6573		pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6574		sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6575
6576		param_len = scsi_2btoul(cdb->length);
6577		header_size = sizeof(struct scsi_mode_header_10);
6578		break;
6579	}
6580	default:
6581		ctl_set_invalid_opcode(ctsio);
6582		ctl_done((union ctl_io *)ctsio);
6583		return (CTL_RETVAL_COMPLETE);
6584		break; /* NOTREACHED */
6585	}
6586
6587	/*
6588	 * From SPC-3:
6589	 * "A parameter list length of zero indicates that the Data-Out Buffer
6590	 * shall be empty. This condition shall not be considered as an error."
6591	 */
6592	if (param_len == 0) {
6593		ctl_set_success(ctsio);
6594		ctl_done((union ctl_io *)ctsio);
6595		return (CTL_RETVAL_COMPLETE);
6596	}
6597
6598	/*
6599	 * Since we'll hit this the first time through, prior to
6600	 * allocation, we don't need to free a data buffer here.
6601	 */
6602	if (param_len < header_size) {
6603		ctl_set_param_len_error(ctsio);
6604		ctl_done((union ctl_io *)ctsio);
6605		return (CTL_RETVAL_COMPLETE);
6606	}
6607
6608	/*
6609	 * Allocate the data buffer and grab the user's data.  In theory,
6610	 * we shouldn't have to sanity check the parameter list length here
6611	 * because the maximum size is 64K.  We should be able to malloc
6612	 * that much without too many problems.
6613	 */
6614	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6615		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6616		ctsio->kern_data_len = param_len;
6617		ctsio->kern_total_len = param_len;
6618		ctsio->kern_data_resid = 0;
6619		ctsio->kern_rel_offset = 0;
6620		ctsio->kern_sg_entries = 0;
6621		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6622		ctsio->be_move_done = ctl_config_move_done;
6623		ctl_datamove((union ctl_io *)ctsio);
6624
6625		return (CTL_RETVAL_COMPLETE);
6626	}
6627
6628	switch (ctsio->cdb[0]) {
6629	case MODE_SELECT_6: {
6630		struct scsi_mode_header_6 *mh6;
6631
6632		mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6633		bd_len = mh6->blk_desc_len;
6634		break;
6635	}
6636	case MODE_SELECT_10: {
6637		struct scsi_mode_header_10 *mh10;
6638
6639		mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6640		bd_len = scsi_2btoul(mh10->blk_desc_len);
6641		break;
6642	}
6643	default:
6644		panic("Invalid CDB type %#x", ctsio->cdb[0]);
6645		break;
6646	}
6647
6648	if (param_len < (header_size + bd_len)) {
6649		free(ctsio->kern_data_ptr, M_CTL);
6650		ctl_set_param_len_error(ctsio);
6651		ctl_done((union ctl_io *)ctsio);
6652		return (CTL_RETVAL_COMPLETE);
6653	}
6654
6655	/*
6656	 * Set the IO_CONT flag, so that if this I/O gets passed to
6657	 * ctl_config_write_done(), it'll get passed back to
6658	 * ctl_do_mode_select() for further processing, or completion if
6659	 * we're all done.
6660	 */
6661	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6662	ctsio->io_cont = ctl_do_mode_select;
6663
6664	modepage_info = (union ctl_modepage_info *)
6665		ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6666
6667	memset(modepage_info, 0, sizeof(*modepage_info));
6668
6669	len_left = param_len - header_size - bd_len;
6670	len_used = header_size + bd_len;
6671
6672	modepage_info->header.len_left = len_left;
6673	modepage_info->header.len_used = len_used;
6674
6675	return (ctl_do_mode_select((union ctl_io *)ctsio));
6676}
6677
6678int
6679ctl_mode_sense(struct ctl_scsiio *ctsio)
6680{
6681	struct ctl_lun *lun;
6682	int pc, page_code, dbd, llba, subpage;
6683	int alloc_len, page_len, header_len, total_len;
6684	struct scsi_mode_block_descr *block_desc;
6685	struct ctl_page_index *page_index;
6686	int control_dev;
6687
6688	dbd = 0;
6689	llba = 0;
6690	block_desc = NULL;
6691	page_index = NULL;
6692
6693	CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6694
6695	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6696
6697	if (lun->be_lun->lun_type != T_DIRECT)
6698		control_dev = 1;
6699	else
6700		control_dev = 0;
6701
6702	switch (ctsio->cdb[0]) {
6703	case MODE_SENSE_6: {
6704		struct scsi_mode_sense_6 *cdb;
6705
6706		cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6707
6708		header_len = sizeof(struct scsi_mode_hdr_6);
6709		if (cdb->byte2 & SMS_DBD)
6710			dbd = 1;
6711		else
6712			header_len += sizeof(struct scsi_mode_block_descr);
6713
6714		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6715		page_code = cdb->page & SMS_PAGE_CODE;
6716		subpage = cdb->subpage;
6717		alloc_len = cdb->length;
6718		break;
6719	}
6720	case MODE_SENSE_10: {
6721		struct scsi_mode_sense_10 *cdb;
6722
6723		cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6724
6725		header_len = sizeof(struct scsi_mode_hdr_10);
6726
6727		if (cdb->byte2 & SMS_DBD)
6728			dbd = 1;
6729		else
6730			header_len += sizeof(struct scsi_mode_block_descr);
6731		if (cdb->byte2 & SMS10_LLBAA)
6732			llba = 1;
6733		pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6734		page_code = cdb->page & SMS_PAGE_CODE;
6735		subpage = cdb->subpage;
6736		alloc_len = scsi_2btoul(cdb->length);
6737		break;
6738	}
6739	default:
6740		ctl_set_invalid_opcode(ctsio);
6741		ctl_done((union ctl_io *)ctsio);
6742		return (CTL_RETVAL_COMPLETE);
6743		break; /* NOTREACHED */
6744	}
6745
6746	/*
6747	 * We have to make a first pass through to calculate the size of
6748	 * the pages that match the user's query.  Then we allocate enough
6749	 * memory to hold it, and actually copy the data into the buffer.
6750	 */
6751	switch (page_code) {
6752	case SMS_ALL_PAGES_PAGE: {
6753		int i;
6754
6755		page_len = 0;
6756
6757		/*
6758		 * At the moment, values other than 0 and 0xff here are
6759		 * reserved according to SPC-3.
6760		 */
6761		if ((subpage != SMS_SUBPAGE_PAGE_0)
6762		 && (subpage != SMS_SUBPAGE_ALL)) {
6763			ctl_set_invalid_field(ctsio,
6764					      /*sks_valid*/ 1,
6765					      /*command*/ 1,
6766					      /*field*/ 3,
6767					      /*bit_valid*/ 0,
6768					      /*bit*/ 0);
6769			ctl_done((union ctl_io *)ctsio);
6770			return (CTL_RETVAL_COMPLETE);
6771		}
6772
6773		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6774			if ((control_dev != 0)
6775			 && (lun->mode_pages.index[i].page_flags &
6776			     CTL_PAGE_FLAG_DISK_ONLY))
6777				continue;
6778
6779			/*
6780			 * We don't use this subpage if the user didn't
6781			 * request all subpages.
6782			 */
6783			if ((lun->mode_pages.index[i].subpage != 0)
6784			 && (subpage == SMS_SUBPAGE_PAGE_0))
6785				continue;
6786
6787#if 0
6788			printf("found page %#x len %d\n",
6789			       lun->mode_pages.index[i].page_code &
6790			       SMPH_PC_MASK,
6791			       lun->mode_pages.index[i].page_len);
6792#endif
6793			page_len += lun->mode_pages.index[i].page_len;
6794		}
6795		break;
6796	}
6797	default: {
6798		int i;
6799
6800		page_len = 0;
6801
6802		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6803			/* Look for the right page code */
6804			if ((lun->mode_pages.index[i].page_code &
6805			     SMPH_PC_MASK) != page_code)
6806				continue;
6807
6808			/* Look for the right subpage or the subpage wildcard*/
6809			if ((lun->mode_pages.index[i].subpage != subpage)
6810			 && (subpage != SMS_SUBPAGE_ALL))
6811				continue;
6812
6813			/* Make sure the page is supported for this dev type */
6814			if ((control_dev != 0)
6815			 && (lun->mode_pages.index[i].page_flags &
6816			     CTL_PAGE_FLAG_DISK_ONLY))
6817				continue;
6818
6819#if 0
6820			printf("found page %#x len %d\n",
6821			       lun->mode_pages.index[i].page_code &
6822			       SMPH_PC_MASK,
6823			       lun->mode_pages.index[i].page_len);
6824#endif
6825
6826			page_len += lun->mode_pages.index[i].page_len;
6827		}
6828
6829		if (page_len == 0) {
6830			ctl_set_invalid_field(ctsio,
6831					      /*sks_valid*/ 1,
6832					      /*command*/ 1,
6833					      /*field*/ 2,
6834					      /*bit_valid*/ 1,
6835					      /*bit*/ 5);
6836			ctl_done((union ctl_io *)ctsio);
6837			return (CTL_RETVAL_COMPLETE);
6838		}
6839		break;
6840	}
6841	}
6842
6843	total_len = header_len + page_len;
6844#if 0
6845	printf("header_len = %d, page_len = %d, total_len = %d\n",
6846	       header_len, page_len, total_len);
6847#endif
6848
6849	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6850	ctsio->kern_sg_entries = 0;
6851	ctsio->kern_data_resid = 0;
6852	ctsio->kern_rel_offset = 0;
6853	if (total_len < alloc_len) {
6854		ctsio->residual = alloc_len - total_len;
6855		ctsio->kern_data_len = total_len;
6856		ctsio->kern_total_len = total_len;
6857	} else {
6858		ctsio->residual = 0;
6859		ctsio->kern_data_len = alloc_len;
6860		ctsio->kern_total_len = alloc_len;
6861	}
6862
6863	switch (ctsio->cdb[0]) {
6864	case MODE_SENSE_6: {
6865		struct scsi_mode_hdr_6 *header;
6866
6867		header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6868
6869		header->datalen = ctl_min(total_len - 1, 254);
6870
6871		if (dbd)
6872			header->block_descr_len = 0;
6873		else
6874			header->block_descr_len =
6875				sizeof(struct scsi_mode_block_descr);
6876		block_desc = (struct scsi_mode_block_descr *)&header[1];
6877		break;
6878	}
6879	case MODE_SENSE_10: {
6880		struct scsi_mode_hdr_10 *header;
6881		int datalen;
6882
6883		header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6884
6885		datalen = ctl_min(total_len - 2, 65533);
6886		scsi_ulto2b(datalen, header->datalen);
6887		if (dbd)
6888			scsi_ulto2b(0, header->block_descr_len);
6889		else
6890			scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6891				    header->block_descr_len);
6892		block_desc = (struct scsi_mode_block_descr *)&header[1];
6893		break;
6894	}
6895	default:
6896		panic("invalid CDB type %#x", ctsio->cdb[0]);
6897		break; /* NOTREACHED */
6898	}
6899
6900	/*
6901	 * If we've got a disk, use its blocksize in the block
6902	 * descriptor.  Otherwise, just set it to 0.
6903	 */
6904	if (dbd == 0) {
6905		if (control_dev != 0)
6906			scsi_ulto3b(lun->be_lun->blocksize,
6907				    block_desc->block_len);
6908		else
6909			scsi_ulto3b(0, block_desc->block_len);
6910	}
6911
6912	switch (page_code) {
6913	case SMS_ALL_PAGES_PAGE: {
6914		int i, data_used;
6915
6916		data_used = header_len;
6917		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6918			struct ctl_page_index *page_index;
6919
6920			page_index = &lun->mode_pages.index[i];
6921
6922			if ((control_dev != 0)
6923			 && (page_index->page_flags &
6924			    CTL_PAGE_FLAG_DISK_ONLY))
6925				continue;
6926
6927			/*
6928			 * We don't use this subpage if the user didn't
6929			 * request all subpages.  We already checked (above)
6930			 * to make sure the user only specified a subpage
6931			 * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6932			 */
6933			if ((page_index->subpage != 0)
6934			 && (subpage == SMS_SUBPAGE_PAGE_0))
6935				continue;
6936
6937			/*
6938			 * Call the handler, if it exists, to update the
6939			 * page to the latest values.
6940			 */
6941			if (page_index->sense_handler != NULL)
6942				page_index->sense_handler(ctsio, page_index,pc);
6943
6944			memcpy(ctsio->kern_data_ptr + data_used,
6945			       page_index->page_data +
6946			       (page_index->page_len * pc),
6947			       page_index->page_len);
6948			data_used += page_index->page_len;
6949		}
6950		break;
6951	}
6952	default: {
6953		int i, data_used;
6954
6955		data_used = header_len;
6956
6957		for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6958			struct ctl_page_index *page_index;
6959
6960			page_index = &lun->mode_pages.index[i];
6961
6962			/* Look for the right page code */
6963			if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6964				continue;
6965
6966			/* Look for the right subpage or the subpage wildcard*/
6967			if ((page_index->subpage != subpage)
6968			 && (subpage != SMS_SUBPAGE_ALL))
6969				continue;
6970
6971			/* Make sure the page is supported for this dev type */
6972			if ((control_dev != 0)
6973			 && (page_index->page_flags &
6974			     CTL_PAGE_FLAG_DISK_ONLY))
6975				continue;
6976
6977			/*
6978			 * Call the handler, if it exists, to update the
6979			 * page to the latest values.
6980			 */
6981			if (page_index->sense_handler != NULL)
6982				page_index->sense_handler(ctsio, page_index,pc);
6983
6984			memcpy(ctsio->kern_data_ptr + data_used,
6985			       page_index->page_data +
6986			       (page_index->page_len * pc),
6987			       page_index->page_len);
6988			data_used += page_index->page_len;
6989		}
6990		break;
6991	}
6992	}
6993
6994	ctsio->scsi_status = SCSI_STATUS_OK;
6995
6996	ctsio->be_move_done = ctl_config_move_done;
6997	ctl_datamove((union ctl_io *)ctsio);
6998
6999	return (CTL_RETVAL_COMPLETE);
7000}
7001
7002int
7003ctl_read_capacity(struct ctl_scsiio *ctsio)
7004{
7005	struct scsi_read_capacity *cdb;
7006	struct scsi_read_capacity_data *data;
7007	struct ctl_lun *lun;
7008	uint32_t lba;
7009
7010	CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7011
7012	cdb = (struct scsi_read_capacity *)ctsio->cdb;
7013
7014	lba = scsi_4btoul(cdb->addr);
7015	if (((cdb->pmi & SRC_PMI) == 0)
7016	 && (lba != 0)) {
7017		ctl_set_invalid_field(/*ctsio*/ ctsio,
7018				      /*sks_valid*/ 1,
7019				      /*command*/ 1,
7020				      /*field*/ 2,
7021				      /*bit_valid*/ 0,
7022				      /*bit*/ 0);
7023		ctl_done((union ctl_io *)ctsio);
7024		return (CTL_RETVAL_COMPLETE);
7025	}
7026
7027	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7028
7029	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7030	data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7031	ctsio->residual = 0;
7032	ctsio->kern_data_len = sizeof(*data);
7033	ctsio->kern_total_len = sizeof(*data);
7034	ctsio->kern_data_resid = 0;
7035	ctsio->kern_rel_offset = 0;
7036	ctsio->kern_sg_entries = 0;
7037
7038	/*
7039	 * If the maximum LBA is greater than 0xfffffffe, the user must
7040	 * issue a SERVICE ACTION IN (16) command, with the read capacity
7041	 * serivce action set.
7042	 */
7043	if (lun->be_lun->maxlba > 0xfffffffe)
7044		scsi_ulto4b(0xffffffff, data->addr);
7045	else
7046		scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7047
7048	/*
7049	 * XXX KDM this may not be 512 bytes...
7050	 */
7051	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7052
7053	ctsio->scsi_status = SCSI_STATUS_OK;
7054
7055	ctsio->be_move_done = ctl_config_move_done;
7056	ctl_datamove((union ctl_io *)ctsio);
7057
7058	return (CTL_RETVAL_COMPLETE);
7059}
7060
7061static int
7062ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7063{
7064	struct scsi_read_capacity_16 *cdb;
7065	struct scsi_read_capacity_data_long *data;
7066	struct ctl_lun *lun;
7067	uint64_t lba;
7068	uint32_t alloc_len;
7069
7070	CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7071
7072	cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7073
7074	alloc_len = scsi_4btoul(cdb->alloc_len);
7075	lba = scsi_8btou64(cdb->addr);
7076
7077	if ((cdb->reladr & SRC16_PMI)
7078	 && (lba != 0)) {
7079		ctl_set_invalid_field(/*ctsio*/ ctsio,
7080				      /*sks_valid*/ 1,
7081				      /*command*/ 1,
7082				      /*field*/ 2,
7083				      /*bit_valid*/ 0,
7084				      /*bit*/ 0);
7085		ctl_done((union ctl_io *)ctsio);
7086		return (CTL_RETVAL_COMPLETE);
7087	}
7088
7089	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7090
7091	ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7092	data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7093
7094	if (sizeof(*data) < alloc_len) {
7095		ctsio->residual = alloc_len - sizeof(*data);
7096		ctsio->kern_data_len = sizeof(*data);
7097		ctsio->kern_total_len = sizeof(*data);
7098	} else {
7099		ctsio->residual = 0;
7100		ctsio->kern_data_len = alloc_len;
7101		ctsio->kern_total_len = alloc_len;
7102	}
7103	ctsio->kern_data_resid = 0;
7104	ctsio->kern_rel_offset = 0;
7105	ctsio->kern_sg_entries = 0;
7106
7107	scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7108	/* XXX KDM this may not be 512 bytes... */
7109	scsi_ulto4b(lun->be_lun->blocksize, data->length);
7110	data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7111	scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7112	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7113		data->lalba_lbp[0] |= SRC16_LBPME;
7114
7115	ctsio->scsi_status = SCSI_STATUS_OK;
7116
7117	ctsio->be_move_done = ctl_config_move_done;
7118	ctl_datamove((union ctl_io *)ctsio);
7119
7120	return (CTL_RETVAL_COMPLETE);
7121}
7122
7123int
7124ctl_service_action_in(struct ctl_scsiio *ctsio)
7125{
7126	struct scsi_service_action_in *cdb;
7127	int retval;
7128
7129	CTL_DEBUG_PRINT(("ctl_service_action_in\n"));
7130
7131	cdb = (struct scsi_service_action_in *)ctsio->cdb;
7132
7133	retval = CTL_RETVAL_COMPLETE;
7134
7135	switch (cdb->service_action) {
7136	case SRC16_SERVICE_ACTION:
7137		retval = ctl_read_capacity_16(ctsio);
7138		break;
7139	default:
7140		ctl_set_invalid_field(/*ctsio*/ ctsio,
7141				      /*sks_valid*/ 1,
7142				      /*command*/ 1,
7143				      /*field*/ 1,
7144				      /*bit_valid*/ 1,
7145				      /*bit*/ 4);
7146		ctl_done((union ctl_io *)ctsio);
7147		break;
7148	}
7149
7150	return (retval);
7151}
7152
7153int
7154ctl_maintenance_in(struct ctl_scsiio *ctsio)
7155{
7156	struct scsi_maintenance_in *cdb;
7157	int retval;
7158	int alloc_len, total_len = 0;
7159	int num_target_port_groups, single;
7160	struct ctl_lun *lun;
7161	struct ctl_softc *softc;
7162	struct scsi_target_group_data *rtg_ptr;
7163	struct scsi_target_port_group_descriptor *tpg_desc_ptr1, *tpg_desc_ptr2;
7164	struct scsi_target_port_descriptor  *tp_desc_ptr1_1, *tp_desc_ptr1_2,
7165	                                    *tp_desc_ptr2_1, *tp_desc_ptr2_2;
7166
7167	CTL_DEBUG_PRINT(("ctl_maintenance_in\n"));
7168
7169	cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7170	softc = control_softc;
7171	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7172
7173	retval = CTL_RETVAL_COMPLETE;
7174
7175	if ((cdb->byte2 & SERVICE_ACTION_MASK) != SA_RPRT_TRGT_GRP) {
7176		ctl_set_invalid_field(/*ctsio*/ ctsio,
7177				      /*sks_valid*/ 1,
7178				      /*command*/ 1,
7179				      /*field*/ 1,
7180				      /*bit_valid*/ 1,
7181				      /*bit*/ 4);
7182		ctl_done((union ctl_io *)ctsio);
7183		return(retval);
7184	}
7185
7186	mtx_lock(&softc->ctl_lock);
7187	single = ctl_is_single;
7188	mtx_unlock(&softc->ctl_lock);
7189
7190	if (single)
7191        	num_target_port_groups = NUM_TARGET_PORT_GROUPS - 1;
7192	else
7193        	num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7194
7195	total_len = sizeof(struct scsi_target_group_data) +
7196		sizeof(struct scsi_target_port_group_descriptor) *
7197		num_target_port_groups +
7198		sizeof(struct scsi_target_port_descriptor) *
7199		NUM_PORTS_PER_GRP * num_target_port_groups;
7200
7201	alloc_len = scsi_4btoul(cdb->length);
7202
7203	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7204
7205	ctsio->kern_sg_entries = 0;
7206
7207	if (total_len < alloc_len) {
7208		ctsio->residual = alloc_len - total_len;
7209		ctsio->kern_data_len = total_len;
7210		ctsio->kern_total_len = total_len;
7211	} else {
7212		ctsio->residual = 0;
7213		ctsio->kern_data_len = alloc_len;
7214		ctsio->kern_total_len = alloc_len;
7215	}
7216	ctsio->kern_data_resid = 0;
7217	ctsio->kern_rel_offset = 0;
7218
7219	rtg_ptr = (struct scsi_target_group_data *)ctsio->kern_data_ptr;
7220
7221	tpg_desc_ptr1 = &rtg_ptr->groups[0];
7222	tp_desc_ptr1_1 = &tpg_desc_ptr1->descriptors[0];
7223	tp_desc_ptr1_2 = (struct scsi_target_port_descriptor *)
7224	        &tp_desc_ptr1_1->desc_list[0];
7225
7226	if (single == 0) {
7227		tpg_desc_ptr2 = (struct scsi_target_port_group_descriptor *)
7228	                &tp_desc_ptr1_2->desc_list[0];
7229		tp_desc_ptr2_1 = &tpg_desc_ptr2->descriptors[0];
7230		tp_desc_ptr2_2 = (struct scsi_target_port_descriptor *)
7231	        	&tp_desc_ptr2_1->desc_list[0];
7232        } else {
7233		tpg_desc_ptr2 = NULL;
7234		tp_desc_ptr2_1 = NULL;
7235		tp_desc_ptr2_2 = NULL;
7236	}
7237
7238	scsi_ulto4b(total_len - 4, rtg_ptr->length);
7239	if (single == 0) {
7240        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7241			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7242				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7243				tpg_desc_ptr2->pref_state =
7244					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7245			} else {
7246				tpg_desc_ptr1->pref_state =
7247					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7248				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7249			}
7250		} else {
7251			if (lun->flags & CTL_LUN_PRIMARY_SC) {
7252				tpg_desc_ptr1->pref_state =
7253					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7254				tpg_desc_ptr2->pref_state = TPG_PRIMARY;
7255			} else {
7256				tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7257				tpg_desc_ptr2->pref_state =
7258					TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7259			}
7260		}
7261	} else {
7262		tpg_desc_ptr1->pref_state = TPG_PRIMARY;
7263	}
7264	tpg_desc_ptr1->support = 0;
7265	tpg_desc_ptr1->target_port_group[1] = 1;
7266	tpg_desc_ptr1->status = TPG_IMPLICIT;
7267	tpg_desc_ptr1->target_port_count= NUM_PORTS_PER_GRP;
7268
7269	if (single == 0) {
7270		tpg_desc_ptr2->support = 0;
7271		tpg_desc_ptr2->target_port_group[1] = 2;
7272		tpg_desc_ptr2->status = TPG_IMPLICIT;
7273		tpg_desc_ptr2->target_port_count = NUM_PORTS_PER_GRP;
7274
7275		tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7276		tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7277
7278		tp_desc_ptr2_1->relative_target_port_identifier[1] = 9;
7279		tp_desc_ptr2_2->relative_target_port_identifier[1] = 10;
7280	} else {
7281        	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS) {
7282			tp_desc_ptr1_1->relative_target_port_identifier[1] = 1;
7283			tp_desc_ptr1_2->relative_target_port_identifier[1] = 2;
7284		} else {
7285			tp_desc_ptr1_1->relative_target_port_identifier[1] = 9;
7286			tp_desc_ptr1_2->relative_target_port_identifier[1] = 10;
7287		}
7288	}
7289
7290	ctsio->be_move_done = ctl_config_move_done;
7291
7292	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7293			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7294			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7295			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7296			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7297
7298	ctl_datamove((union ctl_io *)ctsio);
7299	return(retval);
7300}
7301
7302int
7303ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7304{
7305	struct scsi_per_res_in *cdb;
7306	int alloc_len, total_len = 0;
7307	/* struct scsi_per_res_in_rsrv in_data; */
7308	struct ctl_lun *lun;
7309	struct ctl_softc *softc;
7310
7311	CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7312
7313	softc = control_softc;
7314
7315	cdb = (struct scsi_per_res_in *)ctsio->cdb;
7316
7317	alloc_len = scsi_2btoul(cdb->length);
7318
7319	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7320
7321retry:
7322	mtx_lock(&softc->ctl_lock);
7323	switch (cdb->action) {
7324	case SPRI_RK: /* read keys */
7325		total_len = sizeof(struct scsi_per_res_in_keys) +
7326			lun->pr_key_count *
7327			sizeof(struct scsi_per_res_key);
7328		break;
7329	case SPRI_RR: /* read reservation */
7330		if (lun->flags & CTL_LUN_PR_RESERVED)
7331			total_len = sizeof(struct scsi_per_res_in_rsrv);
7332		else
7333			total_len = sizeof(struct scsi_per_res_in_header);
7334		break;
7335	case SPRI_RC: /* report capabilities */
7336		total_len = sizeof(struct scsi_per_res_cap);
7337		break;
7338	case SPRI_RS: /* read full status */
7339	default:
7340		mtx_unlock(&softc->ctl_lock);
7341		ctl_set_invalid_field(ctsio,
7342				      /*sks_valid*/ 1,
7343				      /*command*/ 1,
7344				      /*field*/ 1,
7345				      /*bit_valid*/ 1,
7346				      /*bit*/ 0);
7347		ctl_done((union ctl_io *)ctsio);
7348		return (CTL_RETVAL_COMPLETE);
7349		break; /* NOTREACHED */
7350	}
7351	mtx_unlock(&softc->ctl_lock);
7352
7353	ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7354
7355	if (total_len < alloc_len) {
7356		ctsio->residual = alloc_len - total_len;
7357		ctsio->kern_data_len = total_len;
7358		ctsio->kern_total_len = total_len;
7359	} else {
7360		ctsio->residual = 0;
7361		ctsio->kern_data_len = alloc_len;
7362		ctsio->kern_total_len = alloc_len;
7363	}
7364
7365	ctsio->kern_data_resid = 0;
7366	ctsio->kern_rel_offset = 0;
7367	ctsio->kern_sg_entries = 0;
7368
7369	mtx_lock(&softc->ctl_lock);
7370	switch (cdb->action) {
7371	case SPRI_RK: { // read keys
7372        struct scsi_per_res_in_keys *res_keys;
7373		int i, key_count;
7374
7375		res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7376
7377		/*
7378		 * We had to drop the lock to allocate our buffer, which
7379		 * leaves time for someone to come in with another
7380		 * persistent reservation.  (That is unlikely, though,
7381		 * since this should be the only persistent reservation
7382		 * command active right now.)
7383		 */
7384		if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7385		    (lun->pr_key_count *
7386		     sizeof(struct scsi_per_res_key)))){
7387			mtx_unlock(&softc->ctl_lock);
7388			free(ctsio->kern_data_ptr, M_CTL);
7389			printf("%s: reservation length changed, retrying\n",
7390			       __func__);
7391			goto retry;
7392		}
7393
7394		scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7395
7396		scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7397			     lun->pr_key_count, res_keys->header.length);
7398
7399		for (i = 0, key_count = 0; i < 2*CTL_MAX_INITIATORS; i++) {
7400			if (!lun->per_res[i].registered)
7401				continue;
7402
7403			/*
7404			 * We used lun->pr_key_count to calculate the
7405			 * size to allocate.  If it turns out the number of
7406			 * initiators with the registered flag set is
7407			 * larger than that (i.e. they haven't been kept in
7408			 * sync), we've got a problem.
7409			 */
7410			if (key_count >= lun->pr_key_count) {
7411#ifdef NEEDTOPORT
7412				csevent_log(CSC_CTL | CSC_SHELF_SW |
7413					    CTL_PR_ERROR,
7414					    csevent_LogType_Fault,
7415					    csevent_AlertLevel_Yellow,
7416					    csevent_FRU_ShelfController,
7417					    csevent_FRU_Firmware,
7418				        csevent_FRU_Unknown,
7419					    "registered keys %d >= key "
7420					    "count %d", key_count,
7421					    lun->pr_key_count);
7422#endif
7423				key_count++;
7424				continue;
7425			}
7426			memcpy(res_keys->keys[key_count].key,
7427			       lun->per_res[i].res_key.key,
7428			       ctl_min(sizeof(res_keys->keys[key_count].key),
7429			       sizeof(lun->per_res[i].res_key)));
7430			key_count++;
7431		}
7432		break;
7433	}
7434	case SPRI_RR: { // read reservation
7435		struct scsi_per_res_in_rsrv *res;
7436		int tmp_len, header_only;
7437
7438		res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7439
7440		scsi_ulto4b(lun->PRGeneration, res->header.generation);
7441
7442		if (lun->flags & CTL_LUN_PR_RESERVED)
7443		{
7444			tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7445			scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7446				    res->header.length);
7447			header_only = 0;
7448		} else {
7449			tmp_len = sizeof(struct scsi_per_res_in_header);
7450			scsi_ulto4b(0, res->header.length);
7451			header_only = 1;
7452		}
7453
7454		/*
7455		 * We had to drop the lock to allocate our buffer, which
7456		 * leaves time for someone to come in with another
7457		 * persistent reservation.  (That is unlikely, though,
7458		 * since this should be the only persistent reservation
7459		 * command active right now.)
7460		 */
7461		if (tmp_len != total_len) {
7462			mtx_unlock(&softc->ctl_lock);
7463			free(ctsio->kern_data_ptr, M_CTL);
7464			printf("%s: reservation status changed, retrying\n",
7465			       __func__);
7466			goto retry;
7467		}
7468
7469		/*
7470		 * No reservation held, so we're done.
7471		 */
7472		if (header_only != 0)
7473			break;
7474
7475		/*
7476		 * If the registration is an All Registrants type, the key
7477		 * is 0, since it doesn't really matter.
7478		 */
7479		if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7480			memcpy(res->data.reservation,
7481			       &lun->per_res[lun->pr_res_idx].res_key,
7482			       sizeof(struct scsi_per_res_key));
7483		}
7484		res->data.scopetype = lun->res_type;
7485		break;
7486	}
7487	case SPRI_RC:     //report capabilities
7488	{
7489		struct scsi_per_res_cap *res_cap;
7490		uint16_t type_mask;
7491
7492		res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7493		scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7494		res_cap->flags2 |= SPRI_TMV;
7495		type_mask = SPRI_TM_WR_EX_AR |
7496			    SPRI_TM_EX_AC_RO |
7497			    SPRI_TM_WR_EX_RO |
7498			    SPRI_TM_EX_AC |
7499			    SPRI_TM_WR_EX |
7500			    SPRI_TM_EX_AC_AR;
7501		scsi_ulto2b(type_mask, res_cap->type_mask);
7502		break;
7503	}
7504	case SPRI_RS: //read full status
7505	default:
7506		/*
7507		 * This is a bug, because we just checked for this above,
7508		 * and should have returned an error.
7509		 */
7510		panic("Invalid PR type %x", cdb->action);
7511		break; /* NOTREACHED */
7512	}
7513	mtx_unlock(&softc->ctl_lock);
7514
7515	ctsio->be_move_done = ctl_config_move_done;
7516
7517	CTL_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n",
7518			 ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1],
7519			 ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3],
7520			 ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5],
7521			 ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7]));
7522
7523	ctl_datamove((union ctl_io *)ctsio);
7524
7525	return (CTL_RETVAL_COMPLETE);
7526}
7527
7528/*
7529 * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7530 * it should return.
7531 */
7532static int
7533ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7534		uint64_t sa_res_key, uint8_t type, uint32_t residx,
7535		struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7536		struct scsi_per_res_out_parms* param)
7537{
7538	union ctl_ha_msg persis_io;
7539	int retval, i;
7540	int isc_retval;
7541
7542	retval = 0;
7543
7544	if (sa_res_key == 0) {
7545		mtx_lock(&softc->ctl_lock);
7546		if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7547			/* validate scope and type */
7548			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7549			     SPR_LU_SCOPE) {
7550				mtx_unlock(&softc->ctl_lock);
7551				ctl_set_invalid_field(/*ctsio*/ ctsio,
7552						      /*sks_valid*/ 1,
7553						      /*command*/ 1,
7554						      /*field*/ 2,
7555						      /*bit_valid*/ 1,
7556						      /*bit*/ 4);
7557				ctl_done((union ctl_io *)ctsio);
7558				return (1);
7559			}
7560
7561		        if (type>8 || type==2 || type==4 || type==0) {
7562				mtx_unlock(&softc->ctl_lock);
7563				ctl_set_invalid_field(/*ctsio*/ ctsio,
7564       	           				      /*sks_valid*/ 1,
7565						      /*command*/ 1,
7566						      /*field*/ 2,
7567						      /*bit_valid*/ 1,
7568						      /*bit*/ 0);
7569				ctl_done((union ctl_io *)ctsio);
7570				return (1);
7571		        }
7572
7573			/* temporarily unregister this nexus */
7574			lun->per_res[residx].registered = 0;
7575
7576			/*
7577			 * Unregister everybody else and build UA for
7578			 * them
7579			 */
7580			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7581				if (lun->per_res[i].registered == 0)
7582					continue;
7583
7584				if (!persis_offset
7585				 && i <CTL_MAX_INITIATORS)
7586					lun->pending_sense[i].ua_pending |=
7587						CTL_UA_REG_PREEMPT;
7588				else if (persis_offset
7589				      && i >= persis_offset)
7590					lun->pending_sense[i-persis_offset
7591						].ua_pending |=
7592						CTL_UA_REG_PREEMPT;
7593				lun->per_res[i].registered = 0;
7594				memset(&lun->per_res[i].res_key, 0,
7595				       sizeof(struct scsi_per_res_key));
7596			}
7597			lun->per_res[residx].registered = 1;
7598			lun->pr_key_count = 1;
7599			lun->res_type = type;
7600			if (lun->res_type != SPR_TYPE_WR_EX_AR
7601			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7602				lun->pr_res_idx = residx;
7603
7604			mtx_unlock(&softc->ctl_lock);
7605			/* send msg to other side */
7606			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7607			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7608			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7609			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7610			persis_io.pr.pr_info.res_type = type;
7611			memcpy(persis_io.pr.pr_info.sa_res_key,
7612			       param->serv_act_res_key,
7613			       sizeof(param->serv_act_res_key));
7614			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7615			     &persis_io, sizeof(persis_io), 0)) >
7616			     CTL_HA_STATUS_SUCCESS) {
7617				printf("CTL:Persis Out error returned "
7618				       "from ctl_ha_msg_send %d\n",
7619				       isc_retval);
7620			}
7621		} else {
7622			/* not all registrants */
7623			mtx_unlock(&softc->ctl_lock);
7624			free(ctsio->kern_data_ptr, M_CTL);
7625			ctl_set_invalid_field(ctsio,
7626					      /*sks_valid*/ 1,
7627					      /*command*/ 0,
7628					      /*field*/ 8,
7629					      /*bit_valid*/ 0,
7630					      /*bit*/ 0);
7631			ctl_done((union ctl_io *)ctsio);
7632			return (1);
7633		}
7634	} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7635		|| !(lun->flags & CTL_LUN_PR_RESERVED)) {
7636		int found = 0;
7637
7638		mtx_lock(&softc->ctl_lock);
7639		if (res_key == sa_res_key) {
7640			/* special case */
7641			/*
7642			 * The spec implies this is not good but doesn't
7643			 * say what to do. There are two choices either
7644			 * generate a res conflict or check condition
7645			 * with illegal field in parameter data. Since
7646			 * that is what is done when the sa_res_key is
7647			 * zero I'll take that approach since this has
7648			 * to do with the sa_res_key.
7649			 */
7650			mtx_unlock(&softc->ctl_lock);
7651			free(ctsio->kern_data_ptr, M_CTL);
7652			ctl_set_invalid_field(ctsio,
7653					      /*sks_valid*/ 1,
7654					      /*command*/ 0,
7655					      /*field*/ 8,
7656					      /*bit_valid*/ 0,
7657					      /*bit*/ 0);
7658			ctl_done((union ctl_io *)ctsio);
7659			return (1);
7660		}
7661
7662		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7663			if (lun->per_res[i].registered
7664			 && memcmp(param->serv_act_res_key,
7665			    lun->per_res[i].res_key.key,
7666			    sizeof(struct scsi_per_res_key)) != 0)
7667				continue;
7668
7669			found = 1;
7670			lun->per_res[i].registered = 0;
7671			memset(&lun->per_res[i].res_key, 0,
7672			       sizeof(struct scsi_per_res_key));
7673			lun->pr_key_count--;
7674
7675			if (!persis_offset
7676			 && i < CTL_MAX_INITIATORS)
7677				lun->pending_sense[i].ua_pending |=
7678					CTL_UA_REG_PREEMPT;
7679			else if (persis_offset
7680			      && i >= persis_offset)
7681				lun->pending_sense[i-persis_offset].ua_pending|=
7682					CTL_UA_REG_PREEMPT;
7683		}
7684		mtx_unlock(&softc->ctl_lock);
7685		if (!found) {
7686			free(ctsio->kern_data_ptr, M_CTL);
7687			ctl_set_reservation_conflict(ctsio);
7688			ctl_done((union ctl_io *)ctsio);
7689			return (CTL_RETVAL_COMPLETE);
7690		}
7691		/* send msg to other side */
7692		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7693		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7694		persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7695		persis_io.pr.pr_info.residx = lun->pr_res_idx;
7696		persis_io.pr.pr_info.res_type = type;
7697		memcpy(persis_io.pr.pr_info.sa_res_key,
7698		       param->serv_act_res_key,
7699		       sizeof(param->serv_act_res_key));
7700		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7701		     &persis_io, sizeof(persis_io), 0)) >
7702		     CTL_HA_STATUS_SUCCESS) {
7703			printf("CTL:Persis Out error returned from "
7704			       "ctl_ha_msg_send %d\n", isc_retval);
7705		}
7706	} else {
7707		/* Reserved but not all registrants */
7708		/* sa_res_key is res holder */
7709		if (memcmp(param->serv_act_res_key,
7710                   lun->per_res[lun->pr_res_idx].res_key.key,
7711                   sizeof(struct scsi_per_res_key)) == 0) {
7712			/* validate scope and type */
7713			if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7714			     SPR_LU_SCOPE) {
7715				ctl_set_invalid_field(/*ctsio*/ ctsio,
7716						      /*sks_valid*/ 1,
7717						      /*command*/ 1,
7718						      /*field*/ 2,
7719						      /*bit_valid*/ 1,
7720						      /*bit*/ 4);
7721				ctl_done((union ctl_io *)ctsio);
7722				return (1);
7723			}
7724
7725			if (type>8 || type==2 || type==4 || type==0) {
7726				ctl_set_invalid_field(/*ctsio*/ ctsio,
7727						      /*sks_valid*/ 1,
7728						      /*command*/ 1,
7729						      /*field*/ 2,
7730						      /*bit_valid*/ 1,
7731						      /*bit*/ 0);
7732				ctl_done((union ctl_io *)ctsio);
7733				return (1);
7734			}
7735
7736			/*
7737			 * Do the following:
7738			 * if sa_res_key != res_key remove all
7739			 * registrants w/sa_res_key and generate UA
7740			 * for these registrants(Registrations
7741			 * Preempted) if it wasn't an exclusive
7742			 * reservation generate UA(Reservations
7743			 * Preempted) for all other registered nexuses
7744			 * if the type has changed. Establish the new
7745			 * reservation and holder. If res_key and
7746			 * sa_res_key are the same do the above
7747			 * except don't unregister the res holder.
7748			 */
7749
7750			/*
7751			 * Temporarily unregister so it won't get
7752			 * removed or UA generated
7753			 */
7754			lun->per_res[residx].registered = 0;
7755			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7756				if (lun->per_res[i].registered == 0)
7757					continue;
7758
7759				if (memcmp(param->serv_act_res_key,
7760				    lun->per_res[i].res_key.key,
7761				    sizeof(struct scsi_per_res_key)) == 0) {
7762					lun->per_res[i].registered = 0;
7763					memset(&lun->per_res[i].res_key,
7764					       0,
7765					       sizeof(struct scsi_per_res_key));
7766					lun->pr_key_count--;
7767
7768					if (!persis_offset
7769					 && i < CTL_MAX_INITIATORS)
7770						lun->pending_sense[i
7771							].ua_pending |=
7772							CTL_UA_REG_PREEMPT;
7773					else if (persis_offset
7774					      && i >= persis_offset)
7775						lun->pending_sense[
7776						  i-persis_offset].ua_pending |=
7777						  CTL_UA_REG_PREEMPT;
7778				} else if (type != lun->res_type
7779					&& (lun->res_type == SPR_TYPE_WR_EX_RO
7780					 || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7781						if (!persis_offset
7782						 && i < CTL_MAX_INITIATORS)
7783							lun->pending_sense[i
7784							].ua_pending |=
7785							CTL_UA_RES_RELEASE;
7786						else if (persis_offset
7787						      && i >= persis_offset)
7788							lun->pending_sense[
7789							i-persis_offset
7790							].ua_pending |=
7791							CTL_UA_RES_RELEASE;
7792				}
7793			}
7794			lun->per_res[residx].registered = 1;
7795			lun->res_type = type;
7796			if (lun->res_type != SPR_TYPE_WR_EX_AR
7797			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7798				lun->pr_res_idx = residx;
7799			else
7800				lun->pr_res_idx =
7801					CTL_PR_ALL_REGISTRANTS;
7802
7803			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7804			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7805			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7806			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7807			persis_io.pr.pr_info.res_type = type;
7808			memcpy(persis_io.pr.pr_info.sa_res_key,
7809			       param->serv_act_res_key,
7810			       sizeof(param->serv_act_res_key));
7811			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7812			     &persis_io, sizeof(persis_io), 0)) >
7813			     CTL_HA_STATUS_SUCCESS) {
7814				printf("CTL:Persis Out error returned "
7815				       "from ctl_ha_msg_send %d\n",
7816				       isc_retval);
7817			}
7818		} else {
7819			/*
7820			 * sa_res_key is not the res holder just
7821			 * remove registrants
7822			 */
7823			int found=0;
7824			mtx_lock(&softc->ctl_lock);
7825
7826			for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7827				if (memcmp(param->serv_act_res_key,
7828				    lun->per_res[i].res_key.key,
7829				    sizeof(struct scsi_per_res_key)) != 0)
7830					continue;
7831
7832				found = 1;
7833				lun->per_res[i].registered = 0;
7834				memset(&lun->per_res[i].res_key, 0,
7835				       sizeof(struct scsi_per_res_key));
7836				lun->pr_key_count--;
7837
7838				if (!persis_offset
7839				 && i < CTL_MAX_INITIATORS)
7840					lun->pending_sense[i].ua_pending |=
7841						CTL_UA_REG_PREEMPT;
7842				else if (persis_offset
7843				      && i >= persis_offset)
7844					lun->pending_sense[
7845						i-persis_offset].ua_pending |=
7846						CTL_UA_REG_PREEMPT;
7847			}
7848
7849			if (!found) {
7850				mtx_unlock(&softc->ctl_lock);
7851				free(ctsio->kern_data_ptr, M_CTL);
7852				ctl_set_reservation_conflict(ctsio);
7853				ctl_done((union ctl_io *)ctsio);
7854		        	return (1);
7855			}
7856			mtx_unlock(&softc->ctl_lock);
7857			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7858			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7859			persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7860			persis_io.pr.pr_info.residx = lun->pr_res_idx;
7861			persis_io.pr.pr_info.res_type = type;
7862			memcpy(persis_io.pr.pr_info.sa_res_key,
7863			       param->serv_act_res_key,
7864			       sizeof(param->serv_act_res_key));
7865			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
7866			     &persis_io, sizeof(persis_io), 0)) >
7867			     CTL_HA_STATUS_SUCCESS) {
7868				printf("CTL:Persis Out error returned "
7869				       "from ctl_ha_msg_send %d\n",
7870				isc_retval);
7871			}
7872		}
7873	}
7874
7875	lun->PRGeneration++;
7876
7877	return (retval);
7878}
7879
7880static void
7881ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7882{
7883	int i;
7884
7885	if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7886	 || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7887	 || memcmp(&lun->per_res[lun->pr_res_idx].res_key,
7888		   msg->pr.pr_info.sa_res_key,
7889		   sizeof(struct scsi_per_res_key)) != 0) {
7890		uint64_t sa_res_key;
7891		sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7892
7893		if (sa_res_key == 0) {
7894			/* temporarily unregister this nexus */
7895			lun->per_res[msg->pr.pr_info.residx].registered = 0;
7896
7897			/*
7898			 * Unregister everybody else and build UA for
7899			 * them
7900			 */
7901			for(i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7902				if (lun->per_res[i].registered == 0)
7903					continue;
7904
7905				if (!persis_offset
7906				 && i < CTL_MAX_INITIATORS)
7907					lun->pending_sense[i].ua_pending |=
7908						CTL_UA_REG_PREEMPT;
7909				else if (persis_offset && i >= persis_offset)
7910					lun->pending_sense[i -
7911						persis_offset].ua_pending |=
7912						CTL_UA_REG_PREEMPT;
7913				lun->per_res[i].registered = 0;
7914				memset(&lun->per_res[i].res_key, 0,
7915				       sizeof(struct scsi_per_res_key));
7916			}
7917
7918			lun->per_res[msg->pr.pr_info.residx].registered = 1;
7919			lun->pr_key_count = 1;
7920			lun->res_type = msg->pr.pr_info.res_type;
7921			if (lun->res_type != SPR_TYPE_WR_EX_AR
7922			 && lun->res_type != SPR_TYPE_EX_AC_AR)
7923				lun->pr_res_idx = msg->pr.pr_info.residx;
7924		} else {
7925		        for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7926				if (memcmp(msg->pr.pr_info.sa_res_key,
7927		                   lun->per_res[i].res_key.key,
7928		                   sizeof(struct scsi_per_res_key)) != 0)
7929					continue;
7930
7931				lun->per_res[i].registered = 0;
7932				memset(&lun->per_res[i].res_key, 0,
7933				       sizeof(struct scsi_per_res_key));
7934				lun->pr_key_count--;
7935
7936				if (!persis_offset
7937				 && i < persis_offset)
7938					lun->pending_sense[i].ua_pending |=
7939						CTL_UA_REG_PREEMPT;
7940				else if (persis_offset
7941				      && i >= persis_offset)
7942					lun->pending_sense[i -
7943						persis_offset].ua_pending |=
7944						CTL_UA_REG_PREEMPT;
7945			}
7946		}
7947	} else {
7948		/*
7949		 * Temporarily unregister so it won't get removed
7950		 * or UA generated
7951		 */
7952		lun->per_res[msg->pr.pr_info.residx].registered = 0;
7953		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
7954			if (lun->per_res[i].registered == 0)
7955				continue;
7956
7957			if (memcmp(msg->pr.pr_info.sa_res_key,
7958	                   lun->per_res[i].res_key.key,
7959	                   sizeof(struct scsi_per_res_key)) == 0) {
7960				lun->per_res[i].registered = 0;
7961				memset(&lun->per_res[i].res_key, 0,
7962				       sizeof(struct scsi_per_res_key));
7963				lun->pr_key_count--;
7964				if (!persis_offset
7965				 && i < CTL_MAX_INITIATORS)
7966					lun->pending_sense[i].ua_pending |=
7967						CTL_UA_REG_PREEMPT;
7968				else if (persis_offset
7969				      && i >= persis_offset)
7970					lun->pending_sense[i -
7971						persis_offset].ua_pending |=
7972						CTL_UA_REG_PREEMPT;
7973			} else if (msg->pr.pr_info.res_type != lun->res_type
7974				&& (lun->res_type == SPR_TYPE_WR_EX_RO
7975				 || lun->res_type == SPR_TYPE_EX_AC_RO)) {
7976					if (!persis_offset
7977					 && i < persis_offset)
7978						lun->pending_sense[i
7979							].ua_pending |=
7980							CTL_UA_RES_RELEASE;
7981					else if (persis_offset
7982					      && i >= persis_offset)
7983					lun->pending_sense[i -
7984						persis_offset].ua_pending |=
7985						CTL_UA_RES_RELEASE;
7986			}
7987		}
7988		lun->per_res[msg->pr.pr_info.residx].registered = 1;
7989		lun->res_type = msg->pr.pr_info.res_type;
7990		if (lun->res_type != SPR_TYPE_WR_EX_AR
7991		 && lun->res_type != SPR_TYPE_EX_AC_AR)
7992			lun->pr_res_idx = msg->pr.pr_info.residx;
7993		else
7994			lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7995	}
7996	lun->PRGeneration++;
7997
7998}
7999
8000
8001int
8002ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8003{
8004	int retval;
8005	int isc_retval;
8006	u_int32_t param_len;
8007	struct scsi_per_res_out *cdb;
8008	struct ctl_lun *lun;
8009	struct scsi_per_res_out_parms* param;
8010	struct ctl_softc *softc;
8011	uint32_t residx;
8012	uint64_t res_key, sa_res_key;
8013	uint8_t type;
8014	union ctl_ha_msg persis_io;
8015	int    i;
8016
8017	CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8018
8019	retval = CTL_RETVAL_COMPLETE;
8020
8021	softc = control_softc;
8022
8023	cdb = (struct scsi_per_res_out *)ctsio->cdb;
8024	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8025
8026	/*
8027	 * We only support whole-LUN scope.  The scope & type are ignored for
8028	 * register, register and ignore existing key and clear.
8029	 * We sometimes ignore scope and type on preempts too!!
8030	 * Verify reservation type here as well.
8031	 */
8032	type = cdb->scope_type & SPR_TYPE_MASK;
8033	if ((cdb->action == SPRO_RESERVE)
8034	 || (cdb->action == SPRO_RELEASE)) {
8035		if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8036			ctl_set_invalid_field(/*ctsio*/ ctsio,
8037					      /*sks_valid*/ 1,
8038					      /*command*/ 1,
8039					      /*field*/ 2,
8040					      /*bit_valid*/ 1,
8041					      /*bit*/ 4);
8042			ctl_done((union ctl_io *)ctsio);
8043			return (CTL_RETVAL_COMPLETE);
8044		}
8045
8046		if (type>8 || type==2 || type==4 || type==0) {
8047			ctl_set_invalid_field(/*ctsio*/ ctsio,
8048					      /*sks_valid*/ 1,
8049					      /*command*/ 1,
8050					      /*field*/ 2,
8051					      /*bit_valid*/ 1,
8052					      /*bit*/ 0);
8053			ctl_done((union ctl_io *)ctsio);
8054			return (CTL_RETVAL_COMPLETE);
8055		}
8056	}
8057
8058	switch (cdb->action & SPRO_ACTION_MASK) {
8059	case SPRO_REGISTER:
8060	case SPRO_RESERVE:
8061	case SPRO_RELEASE:
8062	case SPRO_CLEAR:
8063	case SPRO_PREEMPT:
8064	case SPRO_REG_IGNO:
8065		break;
8066	case SPRO_REG_MOVE:
8067	case SPRO_PRE_ABO:
8068	default:
8069		ctl_set_invalid_field(/*ctsio*/ ctsio,
8070				      /*sks_valid*/ 1,
8071				      /*command*/ 1,
8072				      /*field*/ 1,
8073				      /*bit_valid*/ 1,
8074				      /*bit*/ 0);
8075		ctl_done((union ctl_io *)ctsio);
8076		return (CTL_RETVAL_COMPLETE);
8077		break; /* NOTREACHED */
8078	}
8079
8080	param_len = scsi_4btoul(cdb->length);
8081
8082	if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8083		ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8084		ctsio->kern_data_len = param_len;
8085		ctsio->kern_total_len = param_len;
8086		ctsio->kern_data_resid = 0;
8087		ctsio->kern_rel_offset = 0;
8088		ctsio->kern_sg_entries = 0;
8089		ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8090		ctsio->be_move_done = ctl_config_move_done;
8091		ctl_datamove((union ctl_io *)ctsio);
8092
8093		return (CTL_RETVAL_COMPLETE);
8094	}
8095
8096	param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8097
8098	residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8099	res_key = scsi_8btou64(param->res_key.key);
8100	sa_res_key = scsi_8btou64(param->serv_act_res_key);
8101
8102	/*
8103	 * Validate the reservation key here except for SPRO_REG_IGNO
8104	 * This must be done for all other service actions
8105	 */
8106	if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8107		mtx_lock(&softc->ctl_lock);
8108		if (lun->per_res[residx].registered) {
8109		    if (memcmp(param->res_key.key,
8110			       lun->per_res[residx].res_key.key,
8111			       ctl_min(sizeof(param->res_key),
8112			       sizeof(lun->per_res[residx].res_key))) != 0) {
8113				/*
8114				 * The current key passed in doesn't match
8115				 * the one the initiator previously
8116				 * registered.
8117				 */
8118				mtx_unlock(&softc->ctl_lock);
8119				free(ctsio->kern_data_ptr, M_CTL);
8120				ctl_set_reservation_conflict(ctsio);
8121				ctl_done((union ctl_io *)ctsio);
8122				return (CTL_RETVAL_COMPLETE);
8123			}
8124		} else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8125			/*
8126			 * We are not registered
8127			 */
8128			mtx_unlock(&softc->ctl_lock);
8129			free(ctsio->kern_data_ptr, M_CTL);
8130			ctl_set_reservation_conflict(ctsio);
8131			ctl_done((union ctl_io *)ctsio);
8132			return (CTL_RETVAL_COMPLETE);
8133		} else if (res_key != 0) {
8134			/*
8135			 * We are not registered and trying to register but
8136			 * the register key isn't zero.
8137			 */
8138			mtx_unlock(&softc->ctl_lock);
8139			free(ctsio->kern_data_ptr, M_CTL);
8140			ctl_set_reservation_conflict(ctsio);
8141			ctl_done((union ctl_io *)ctsio);
8142			return (CTL_RETVAL_COMPLETE);
8143		}
8144		mtx_unlock(&softc->ctl_lock);
8145	}
8146
8147	switch (cdb->action & SPRO_ACTION_MASK) {
8148	case SPRO_REGISTER:
8149	case SPRO_REG_IGNO: {
8150
8151#if 0
8152		printf("Registration received\n");
8153#endif
8154
8155		/*
8156		 * We don't support any of these options, as we report in
8157		 * the read capabilities request (see
8158		 * ctl_persistent_reserve_in(), above).
8159		 */
8160		if ((param->flags & SPR_SPEC_I_PT)
8161		 || (param->flags & SPR_ALL_TG_PT)
8162		 || (param->flags & SPR_APTPL)) {
8163			int bit_ptr;
8164
8165			if (param->flags & SPR_APTPL)
8166				bit_ptr = 0;
8167			else if (param->flags & SPR_ALL_TG_PT)
8168				bit_ptr = 2;
8169			else /* SPR_SPEC_I_PT */
8170				bit_ptr = 3;
8171
8172			free(ctsio->kern_data_ptr, M_CTL);
8173			ctl_set_invalid_field(ctsio,
8174					      /*sks_valid*/ 1,
8175					      /*command*/ 0,
8176					      /*field*/ 20,
8177					      /*bit_valid*/ 1,
8178					      /*bit*/ bit_ptr);
8179			ctl_done((union ctl_io *)ctsio);
8180			return (CTL_RETVAL_COMPLETE);
8181		}
8182
8183		mtx_lock(&softc->ctl_lock);
8184
8185		/*
8186		 * The initiator wants to clear the
8187		 * key/unregister.
8188		 */
8189		if (sa_res_key == 0) {
8190			if ((res_key == 0
8191			  && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8192			 || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8193			  && !lun->per_res[residx].registered)) {
8194				mtx_unlock(&softc->ctl_lock);
8195				goto done;
8196			}
8197
8198			lun->per_res[residx].registered = 0;
8199			memset(&lun->per_res[residx].res_key,
8200			       0, sizeof(lun->per_res[residx].res_key));
8201			lun->pr_key_count--;
8202
8203			if (residx == lun->pr_res_idx) {
8204				lun->flags &= ~CTL_LUN_PR_RESERVED;
8205				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8206
8207				if ((lun->res_type == SPR_TYPE_WR_EX_RO
8208				  || lun->res_type == SPR_TYPE_EX_AC_RO)
8209				 && lun->pr_key_count) {
8210					/*
8211					 * If the reservation is a registrants
8212					 * only type we need to generate a UA
8213					 * for other registered inits.  The
8214					 * sense code should be RESERVATIONS
8215					 * RELEASED
8216					 */
8217
8218					for (i = 0; i < CTL_MAX_INITIATORS;i++){
8219						if (lun->per_res[
8220						    i+persis_offset].registered
8221						    == 0)
8222							continue;
8223						lun->pending_sense[i
8224							].ua_pending |=
8225							CTL_UA_RES_RELEASE;
8226					}
8227				}
8228				lun->res_type = 0;
8229			} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8230				if (lun->pr_key_count==0) {
8231					lun->flags &= ~CTL_LUN_PR_RESERVED;
8232					lun->res_type = 0;
8233					lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8234				}
8235			}
8236			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8237			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8238			persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8239			persis_io.pr.pr_info.residx = residx;
8240			if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8241			     &persis_io, sizeof(persis_io), 0 )) >
8242			     CTL_HA_STATUS_SUCCESS) {
8243				printf("CTL:Persis Out error returned from "
8244				       "ctl_ha_msg_send %d\n", isc_retval);
8245			}
8246			mtx_unlock(&softc->ctl_lock);
8247		} else /* sa_res_key != 0 */ {
8248
8249			/*
8250			 * If we aren't registered currently then increment
8251			 * the key count and set the registered flag.
8252			 */
8253			if (!lun->per_res[residx].registered) {
8254				lun->pr_key_count++;
8255				lun->per_res[residx].registered = 1;
8256			}
8257
8258			memcpy(&lun->per_res[residx].res_key,
8259			       param->serv_act_res_key,
8260			       ctl_min(sizeof(param->serv_act_res_key),
8261			       sizeof(lun->per_res[residx].res_key)));
8262
8263			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8264			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8265			persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8266			persis_io.pr.pr_info.residx = residx;
8267			memcpy(persis_io.pr.pr_info.sa_res_key,
8268			       param->serv_act_res_key,
8269			       sizeof(param->serv_act_res_key));
8270			mtx_unlock(&softc->ctl_lock);
8271			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8272			     &persis_io, sizeof(persis_io), 0)) >
8273			     CTL_HA_STATUS_SUCCESS) {
8274				printf("CTL:Persis Out error returned from "
8275				       "ctl_ha_msg_send %d\n", isc_retval);
8276			}
8277		}
8278		lun->PRGeneration++;
8279
8280		break;
8281	}
8282	case SPRO_RESERVE:
8283#if 0
8284                printf("Reserve executed type %d\n", type);
8285#endif
8286		mtx_lock(&softc->ctl_lock);
8287		if (lun->flags & CTL_LUN_PR_RESERVED) {
8288			/*
8289			 * if this isn't the reservation holder and it's
8290			 * not a "all registrants" type or if the type is
8291			 * different then we have a conflict
8292			 */
8293			if ((lun->pr_res_idx != residx
8294			  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8295			 || lun->res_type != type) {
8296				mtx_unlock(&softc->ctl_lock);
8297				free(ctsio->kern_data_ptr, M_CTL);
8298				ctl_set_reservation_conflict(ctsio);
8299				ctl_done((union ctl_io *)ctsio);
8300				return (CTL_RETVAL_COMPLETE);
8301			}
8302			mtx_unlock(&softc->ctl_lock);
8303		} else /* create a reservation */ {
8304			/*
8305			 * If it's not an "all registrants" type record
8306			 * reservation holder
8307			 */
8308			if (type != SPR_TYPE_WR_EX_AR
8309			 && type != SPR_TYPE_EX_AC_AR)
8310				lun->pr_res_idx = residx; /* Res holder */
8311			else
8312				lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8313
8314			lun->flags |= CTL_LUN_PR_RESERVED;
8315			lun->res_type = type;
8316
8317			mtx_unlock(&softc->ctl_lock);
8318
8319			/* send msg to other side */
8320			persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8321			persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8322			persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8323			persis_io.pr.pr_info.residx = lun->pr_res_idx;
8324			persis_io.pr.pr_info.res_type = type;
8325			if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
8326			     &persis_io, sizeof(persis_io), 0)) >
8327			     CTL_HA_STATUS_SUCCESS) {
8328				printf("CTL:Persis Out error returned from "
8329				       "ctl_ha_msg_send %d\n", isc_retval);
8330			}
8331		}
8332		break;
8333
8334	case SPRO_RELEASE:
8335		mtx_lock(&softc->ctl_lock);
8336		if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8337			/* No reservation exists return good status */
8338			mtx_unlock(&softc->ctl_lock);
8339			goto done;
8340		}
8341		/*
8342		 * Is this nexus a reservation holder?
8343		 */
8344		if (lun->pr_res_idx != residx
8345		 && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8346			/*
8347			 * not a res holder return good status but
8348			 * do nothing
8349			 */
8350			mtx_unlock(&softc->ctl_lock);
8351			goto done;
8352		}
8353
8354		if (lun->res_type != type) {
8355			mtx_unlock(&softc->ctl_lock);
8356			free(ctsio->kern_data_ptr, M_CTL);
8357			ctl_set_illegal_pr_release(ctsio);
8358			ctl_done((union ctl_io *)ctsio);
8359			return (CTL_RETVAL_COMPLETE);
8360		}
8361
8362		/* okay to release */
8363		lun->flags &= ~CTL_LUN_PR_RESERVED;
8364		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8365		lun->res_type = 0;
8366
8367		/*
8368		 * if this isn't an exclusive access
8369		 * res generate UA for all other
8370		 * registrants.
8371		 */
8372		if (type != SPR_TYPE_EX_AC
8373		 && type != SPR_TYPE_WR_EX) {
8374			/*
8375			 * temporarily unregister so we don't generate UA
8376			 */
8377			lun->per_res[residx].registered = 0;
8378
8379			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8380				if (lun->per_res[i+persis_offset].registered
8381				    == 0)
8382					continue;
8383				lun->pending_sense[i].ua_pending |=
8384					CTL_UA_RES_RELEASE;
8385			}
8386
8387			lun->per_res[residx].registered = 1;
8388		}
8389		mtx_unlock(&softc->ctl_lock);
8390		/* Send msg to other side */
8391		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8392		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8393		persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8394		if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io,
8395		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8396			printf("CTL:Persis Out error returned from "
8397			       "ctl_ha_msg_send %d\n", isc_retval);
8398		}
8399		break;
8400
8401	case SPRO_CLEAR:
8402		/* send msg to other side */
8403
8404		mtx_lock(&softc->ctl_lock);
8405		lun->flags &= ~CTL_LUN_PR_RESERVED;
8406		lun->res_type = 0;
8407		lun->pr_key_count = 0;
8408		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8409
8410
8411		memset(&lun->per_res[residx].res_key,
8412		       0, sizeof(lun->per_res[residx].res_key));
8413		lun->per_res[residx].registered = 0;
8414
8415		for (i=0; i < 2*CTL_MAX_INITIATORS; i++)
8416			if (lun->per_res[i].registered) {
8417				if (!persis_offset && i < CTL_MAX_INITIATORS)
8418					lun->pending_sense[i].ua_pending |=
8419						CTL_UA_RES_PREEMPT;
8420				else if (persis_offset && i >= persis_offset)
8421					lun->pending_sense[i-persis_offset
8422					    ].ua_pending |= CTL_UA_RES_PREEMPT;
8423
8424				memset(&lun->per_res[i].res_key,
8425				       0, sizeof(struct scsi_per_res_key));
8426				lun->per_res[i].registered = 0;
8427			}
8428		lun->PRGeneration++;
8429		mtx_unlock(&softc->ctl_lock);
8430		persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8431		persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8432		persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8433		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8434		     sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) {
8435			printf("CTL:Persis Out error returned from "
8436			       "ctl_ha_msg_send %d\n", isc_retval);
8437		}
8438		break;
8439
8440	case SPRO_PREEMPT: {
8441		int nretval;
8442
8443		nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8444					  residx, ctsio, cdb, param);
8445		if (nretval != 0)
8446			return (CTL_RETVAL_COMPLETE);
8447		break;
8448	}
8449	case SPRO_REG_MOVE:
8450	case SPRO_PRE_ABO:
8451	default:
8452		free(ctsio->kern_data_ptr, M_CTL);
8453		ctl_set_invalid_field(/*ctsio*/ ctsio,
8454				      /*sks_valid*/ 1,
8455				      /*command*/ 1,
8456				      /*field*/ 1,
8457				      /*bit_valid*/ 1,
8458				      /*bit*/ 0);
8459		ctl_done((union ctl_io *)ctsio);
8460		return (CTL_RETVAL_COMPLETE);
8461		break; /* NOTREACHED */
8462	}
8463
8464done:
8465	free(ctsio->kern_data_ptr, M_CTL);
8466	ctl_set_success(ctsio);
8467	ctl_done((union ctl_io *)ctsio);
8468
8469	return (retval);
8470}
8471
8472/*
8473 * This routine is for handling a message from the other SC pertaining to
8474 * persistent reserve out. All the error checking will have been done
8475 * so only perorming the action need be done here to keep the two
8476 * in sync.
8477 */
8478static void
8479ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8480{
8481	struct ctl_lun *lun;
8482	struct ctl_softc *softc;
8483	int i;
8484	uint32_t targ_lun;
8485
8486	softc = control_softc;
8487
8488	mtx_lock(&softc->ctl_lock);
8489
8490	targ_lun = msg->hdr.nexus.targ_lun;
8491	if (msg->hdr.nexus.lun_map_fn != NULL)
8492		targ_lun = msg->hdr.nexus.lun_map_fn(msg->hdr.nexus.lun_map_arg, targ_lun);
8493	lun = softc->ctl_luns[targ_lun];
8494	switch(msg->pr.pr_info.action) {
8495	case CTL_PR_REG_KEY:
8496		if (!lun->per_res[msg->pr.pr_info.residx].registered) {
8497			lun->per_res[msg->pr.pr_info.residx].registered = 1;
8498			lun->pr_key_count++;
8499		}
8500		lun->PRGeneration++;
8501		memcpy(&lun->per_res[msg->pr.pr_info.residx].res_key,
8502		       msg->pr.pr_info.sa_res_key,
8503		       sizeof(struct scsi_per_res_key));
8504		break;
8505
8506	case CTL_PR_UNREG_KEY:
8507		lun->per_res[msg->pr.pr_info.residx].registered = 0;
8508		memset(&lun->per_res[msg->pr.pr_info.residx].res_key,
8509		       0, sizeof(struct scsi_per_res_key));
8510		lun->pr_key_count--;
8511
8512		/* XXX Need to see if the reservation has been released */
8513		/* if so do we need to generate UA? */
8514		if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8515			lun->flags &= ~CTL_LUN_PR_RESERVED;
8516			lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8517
8518			if ((lun->res_type == SPR_TYPE_WR_EX_RO
8519			  || lun->res_type == SPR_TYPE_EX_AC_RO)
8520			 && lun->pr_key_count) {
8521				/*
8522				 * If the reservation is a registrants
8523				 * only type we need to generate a UA
8524				 * for other registered inits.  The
8525				 * sense code should be RESERVATIONS
8526				 * RELEASED
8527				 */
8528
8529				for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8530					if (lun->per_res[i+
8531					    persis_offset].registered == 0)
8532						continue;
8533
8534					lun->pending_sense[i
8535						].ua_pending |=
8536						CTL_UA_RES_RELEASE;
8537				}
8538			}
8539			lun->res_type = 0;
8540		} else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8541			if (lun->pr_key_count==0) {
8542				lun->flags &= ~CTL_LUN_PR_RESERVED;
8543				lun->res_type = 0;
8544				lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8545			}
8546		}
8547		lun->PRGeneration++;
8548		break;
8549
8550	case CTL_PR_RESERVE:
8551		lun->flags |= CTL_LUN_PR_RESERVED;
8552		lun->res_type = msg->pr.pr_info.res_type;
8553		lun->pr_res_idx = msg->pr.pr_info.residx;
8554
8555		break;
8556
8557	case CTL_PR_RELEASE:
8558		/*
8559		 * if this isn't an exclusive access res generate UA for all
8560		 * other registrants.
8561		 */
8562		if (lun->res_type != SPR_TYPE_EX_AC
8563		 && lun->res_type != SPR_TYPE_WR_EX) {
8564			for (i = 0; i < CTL_MAX_INITIATORS; i++)
8565				if (lun->per_res[i+persis_offset].registered)
8566					lun->pending_sense[i].ua_pending |=
8567						CTL_UA_RES_RELEASE;
8568		}
8569
8570		lun->flags &= ~CTL_LUN_PR_RESERVED;
8571		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8572		lun->res_type = 0;
8573		break;
8574
8575	case CTL_PR_PREEMPT:
8576		ctl_pro_preempt_other(lun, msg);
8577		break;
8578	case CTL_PR_CLEAR:
8579		lun->flags &= ~CTL_LUN_PR_RESERVED;
8580		lun->res_type = 0;
8581		lun->pr_key_count = 0;
8582		lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8583
8584		for (i=0; i < 2*CTL_MAX_INITIATORS; i++) {
8585			if (lun->per_res[i].registered == 0)
8586				continue;
8587			if (!persis_offset
8588			 && i < CTL_MAX_INITIATORS)
8589				lun->pending_sense[i].ua_pending |=
8590					CTL_UA_RES_PREEMPT;
8591			else if (persis_offset
8592			      && i >= persis_offset)
8593   				lun->pending_sense[i-persis_offset].ua_pending|=
8594					CTL_UA_RES_PREEMPT;
8595			memset(&lun->per_res[i].res_key, 0,
8596			       sizeof(struct scsi_per_res_key));
8597			lun->per_res[i].registered = 0;
8598		}
8599		lun->PRGeneration++;
8600		break;
8601	}
8602
8603	mtx_unlock(&softc->ctl_lock);
8604}
8605
8606int
8607ctl_read_write(struct ctl_scsiio *ctsio)
8608{
8609	struct ctl_lun *lun;
8610	struct ctl_lba_len_flags *lbalen;
8611	uint64_t lba;
8612	uint32_t num_blocks;
8613	int reladdr, fua, dpo, ebp;
8614	int retval;
8615	int isread;
8616
8617	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8618
8619	CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8620
8621	reladdr = 0;
8622	fua = 0;
8623	dpo = 0;
8624	ebp = 0;
8625
8626	retval = CTL_RETVAL_COMPLETE;
8627
8628	isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8629	      || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8630	if (lun->flags & CTL_LUN_PR_RESERVED && isread) {
8631		uint32_t residx;
8632
8633		/*
8634		 * XXX KDM need a lock here.
8635		 */
8636		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
8637		if ((lun->res_type == SPR_TYPE_EX_AC
8638		  && residx != lun->pr_res_idx)
8639		 || ((lun->res_type == SPR_TYPE_EX_AC_RO
8640		   || lun->res_type == SPR_TYPE_EX_AC_AR)
8641		  && !lun->per_res[residx].registered)) {
8642			ctl_set_reservation_conflict(ctsio);
8643			ctl_done((union ctl_io *)ctsio);
8644			return (CTL_RETVAL_COMPLETE);
8645	        }
8646	}
8647
8648	switch (ctsio->cdb[0]) {
8649	case READ_6:
8650	case WRITE_6: {
8651		struct scsi_rw_6 *cdb;
8652
8653		cdb = (struct scsi_rw_6 *)ctsio->cdb;
8654
8655		lba = scsi_3btoul(cdb->addr);
8656		/* only 5 bits are valid in the most significant address byte */
8657		lba &= 0x1fffff;
8658		num_blocks = cdb->length;
8659		/*
8660		 * This is correct according to SBC-2.
8661		 */
8662		if (num_blocks == 0)
8663			num_blocks = 256;
8664		break;
8665	}
8666	case READ_10:
8667	case WRITE_10: {
8668		struct scsi_rw_10 *cdb;
8669
8670		cdb = (struct scsi_rw_10 *)ctsio->cdb;
8671
8672		if (cdb->byte2 & SRW10_RELADDR)
8673			reladdr = 1;
8674		if (cdb->byte2 & SRW10_FUA)
8675			fua = 1;
8676		if (cdb->byte2 & SRW10_DPO)
8677			dpo = 1;
8678
8679		if ((cdb->opcode == WRITE_10)
8680		 && (cdb->byte2 & SRW10_EBP))
8681			ebp = 1;
8682
8683		lba = scsi_4btoul(cdb->addr);
8684		num_blocks = scsi_2btoul(cdb->length);
8685		break;
8686	}
8687	case WRITE_VERIFY_10: {
8688		struct scsi_write_verify_10 *cdb;
8689
8690		cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8691
8692		/*
8693		 * XXX KDM we should do actual write verify support at some
8694		 * point.  This is obviously fake, we're just translating
8695		 * things to a write.  So we don't even bother checking the
8696		 * BYTCHK field, since we don't do any verification.  If
8697		 * the user asks for it, we'll just pretend we did it.
8698		 */
8699		if (cdb->byte2 & SWV_DPO)
8700			dpo = 1;
8701
8702		lba = scsi_4btoul(cdb->addr);
8703		num_blocks = scsi_2btoul(cdb->length);
8704		break;
8705	}
8706	case READ_12:
8707	case WRITE_12: {
8708		struct scsi_rw_12 *cdb;
8709
8710		cdb = (struct scsi_rw_12 *)ctsio->cdb;
8711
8712		if (cdb->byte2 & SRW12_RELADDR)
8713			reladdr = 1;
8714		if (cdb->byte2 & SRW12_FUA)
8715			fua = 1;
8716		if (cdb->byte2 & SRW12_DPO)
8717			dpo = 1;
8718		lba = scsi_4btoul(cdb->addr);
8719		num_blocks = scsi_4btoul(cdb->length);
8720		break;
8721	}
8722	case WRITE_VERIFY_12: {
8723		struct scsi_write_verify_12 *cdb;
8724
8725		cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8726
8727		if (cdb->byte2 & SWV_DPO)
8728			dpo = 1;
8729
8730		lba = scsi_4btoul(cdb->addr);
8731		num_blocks = scsi_4btoul(cdb->length);
8732
8733		break;
8734	}
8735	case READ_16:
8736	case WRITE_16: {
8737		struct scsi_rw_16 *cdb;
8738
8739		cdb = (struct scsi_rw_16 *)ctsio->cdb;
8740
8741		if (cdb->byte2 & SRW12_RELADDR)
8742			reladdr = 1;
8743		if (cdb->byte2 & SRW12_FUA)
8744			fua = 1;
8745		if (cdb->byte2 & SRW12_DPO)
8746			dpo = 1;
8747
8748		lba = scsi_8btou64(cdb->addr);
8749		num_blocks = scsi_4btoul(cdb->length);
8750		break;
8751	}
8752	case WRITE_VERIFY_16: {
8753		struct scsi_write_verify_16 *cdb;
8754
8755		cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8756
8757		if (cdb->byte2 & SWV_DPO)
8758			dpo = 1;
8759
8760		lba = scsi_8btou64(cdb->addr);
8761		num_blocks = scsi_4btoul(cdb->length);
8762		break;
8763	}
8764	default:
8765		/*
8766		 * We got a command we don't support.  This shouldn't
8767		 * happen, commands should be filtered out above us.
8768		 */
8769		ctl_set_invalid_opcode(ctsio);
8770		ctl_done((union ctl_io *)ctsio);
8771
8772		return (CTL_RETVAL_COMPLETE);
8773		break; /* NOTREACHED */
8774	}
8775
8776	/*
8777	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8778	 * interesting for us, but if RAIDCore is in write-back mode,
8779	 * getting it to do write-through for a particular transaction may
8780	 * not be possible.
8781	 */
8782	/*
8783	 * We don't support relative addressing.  That also requires
8784	 * supporting linked commands, which we don't do.
8785	 */
8786	if (reladdr != 0) {
8787		ctl_set_invalid_field(ctsio,
8788				      /*sks_valid*/ 1,
8789				      /*command*/ 1,
8790				      /*field*/ 1,
8791				      /*bit_valid*/ 1,
8792				      /*bit*/ 0);
8793		ctl_done((union ctl_io *)ctsio);
8794		return (CTL_RETVAL_COMPLETE);
8795	}
8796
8797	/*
8798	 * The first check is to make sure we're in bounds, the second
8799	 * check is to catch wrap-around problems.  If the lba + num blocks
8800	 * is less than the lba, then we've wrapped around and the block
8801	 * range is invalid anyway.
8802	 */
8803	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8804	 || ((lba + num_blocks) < lba)) {
8805		ctl_set_lba_out_of_range(ctsio);
8806		ctl_done((union ctl_io *)ctsio);
8807		return (CTL_RETVAL_COMPLETE);
8808	}
8809
8810	/*
8811	 * According to SBC-3, a transfer length of 0 is not an error.
8812	 * Note that this cannot happen with WRITE(6) or READ(6), since 0
8813	 * translates to 256 blocks for those commands.
8814	 */
8815	if (num_blocks == 0) {
8816		ctl_set_success(ctsio);
8817		ctl_done((union ctl_io *)ctsio);
8818		return (CTL_RETVAL_COMPLETE);
8819	}
8820
8821	lbalen = (struct ctl_lba_len_flags *)
8822	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8823	lbalen->lba = lba;
8824	lbalen->len = num_blocks;
8825	lbalen->flags = isread ? CTL_LLF_READ : CTL_LLF_WRITE;
8826
8827	ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8828	ctsio->kern_rel_offset = 0;
8829
8830	CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8831
8832	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8833
8834	return (retval);
8835}
8836
8837static int
8838ctl_cnw_cont(union ctl_io *io)
8839{
8840	struct ctl_scsiio *ctsio;
8841	struct ctl_lun *lun;
8842	struct ctl_lba_len_flags *lbalen;
8843	int retval;
8844
8845	ctsio = &io->scsiio;
8846	ctsio->io_hdr.status = CTL_STATUS_NONE;
8847	ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8848	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8849	lbalen = (struct ctl_lba_len_flags *)
8850	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8851	lbalen->flags = CTL_LLF_WRITE;
8852
8853	CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8854	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8855	return (retval);
8856}
8857
8858int
8859ctl_cnw(struct ctl_scsiio *ctsio)
8860{
8861	struct ctl_lun *lun;
8862	struct ctl_lba_len_flags *lbalen;
8863	uint64_t lba;
8864	uint32_t num_blocks;
8865	int fua, dpo;
8866	int retval;
8867
8868	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8869
8870	CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8871
8872	fua = 0;
8873	dpo = 0;
8874
8875	retval = CTL_RETVAL_COMPLETE;
8876
8877	switch (ctsio->cdb[0]) {
8878	case COMPARE_AND_WRITE: {
8879		struct scsi_compare_and_write *cdb;
8880
8881		cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8882
8883		if (cdb->byte2 & SRW10_FUA)
8884			fua = 1;
8885		if (cdb->byte2 & SRW10_DPO)
8886			dpo = 1;
8887		lba = scsi_8btou64(cdb->addr);
8888		num_blocks = cdb->length;
8889		break;
8890	}
8891	default:
8892		/*
8893		 * We got a command we don't support.  This shouldn't
8894		 * happen, commands should be filtered out above us.
8895		 */
8896		ctl_set_invalid_opcode(ctsio);
8897		ctl_done((union ctl_io *)ctsio);
8898
8899		return (CTL_RETVAL_COMPLETE);
8900		break; /* NOTREACHED */
8901	}
8902
8903	/*
8904	 * XXX KDM what do we do with the DPO and FUA bits?  FUA might be
8905	 * interesting for us, but if RAIDCore is in write-back mode,
8906	 * getting it to do write-through for a particular transaction may
8907	 * not be possible.
8908	 */
8909
8910	/*
8911	 * The first check is to make sure we're in bounds, the second
8912	 * check is to catch wrap-around problems.  If the lba + num blocks
8913	 * is less than the lba, then we've wrapped around and the block
8914	 * range is invalid anyway.
8915	 */
8916	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8917	 || ((lba + num_blocks) < lba)) {
8918		ctl_set_lba_out_of_range(ctsio);
8919		ctl_done((union ctl_io *)ctsio);
8920		return (CTL_RETVAL_COMPLETE);
8921	}
8922
8923	/*
8924	 * According to SBC-3, a transfer length of 0 is not an error.
8925	 */
8926	if (num_blocks == 0) {
8927		ctl_set_success(ctsio);
8928		ctl_done((union ctl_io *)ctsio);
8929		return (CTL_RETVAL_COMPLETE);
8930	}
8931
8932	ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8933	ctsio->kern_rel_offset = 0;
8934
8935	/*
8936	 * Set the IO_CONT flag, so that if this I/O gets passed to
8937	 * ctl_data_submit_done(), it'll get passed back to
8938	 * ctl_ctl_cnw_cont() for further processing.
8939	 */
8940	ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8941	ctsio->io_cont = ctl_cnw_cont;
8942
8943	lbalen = (struct ctl_lba_len_flags *)
8944	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8945	lbalen->lba = lba;
8946	lbalen->len = num_blocks;
8947	lbalen->flags = CTL_LLF_COMPARE;
8948
8949	CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8950	retval = lun->backend->data_submit((union ctl_io *)ctsio);
8951	return (retval);
8952}
8953
8954int
8955ctl_verify(struct ctl_scsiio *ctsio)
8956{
8957	struct ctl_lun *lun;
8958	struct ctl_lba_len_flags *lbalen;
8959	uint64_t lba;
8960	uint32_t num_blocks;
8961	int bytchk, dpo;
8962	int retval;
8963
8964	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8965
8966	CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8967
8968	bytchk = 0;
8969	dpo = 0;
8970	retval = CTL_RETVAL_COMPLETE;
8971
8972	switch (ctsio->cdb[0]) {
8973	case VERIFY_10: {
8974		struct scsi_verify_10 *cdb;
8975
8976		cdb = (struct scsi_verify_10 *)ctsio->cdb;
8977		if (cdb->byte2 & SVFY_BYTCHK)
8978			bytchk = 1;
8979		if (cdb->byte2 & SVFY_DPO)
8980			dpo = 1;
8981		lba = scsi_4btoul(cdb->addr);
8982		num_blocks = scsi_2btoul(cdb->length);
8983		break;
8984	}
8985	case VERIFY_12: {
8986		struct scsi_verify_12 *cdb;
8987
8988		cdb = (struct scsi_verify_12 *)ctsio->cdb;
8989		if (cdb->byte2 & SVFY_BYTCHK)
8990			bytchk = 1;
8991		if (cdb->byte2 & SVFY_DPO)
8992			dpo = 1;
8993		lba = scsi_4btoul(cdb->addr);
8994		num_blocks = scsi_4btoul(cdb->length);
8995		break;
8996	}
8997	case VERIFY_16: {
8998		struct scsi_rw_16 *cdb;
8999
9000		cdb = (struct scsi_rw_16 *)ctsio->cdb;
9001		if (cdb->byte2 & SVFY_BYTCHK)
9002			bytchk = 1;
9003		if (cdb->byte2 & SVFY_DPO)
9004			dpo = 1;
9005		lba = scsi_8btou64(cdb->addr);
9006		num_blocks = scsi_4btoul(cdb->length);
9007		break;
9008	}
9009	default:
9010		/*
9011		 * We got a command we don't support.  This shouldn't
9012		 * happen, commands should be filtered out above us.
9013		 */
9014		ctl_set_invalid_opcode(ctsio);
9015		ctl_done((union ctl_io *)ctsio);
9016		return (CTL_RETVAL_COMPLETE);
9017	}
9018
9019	/*
9020	 * The first check is to make sure we're in bounds, the second
9021	 * check is to catch wrap-around problems.  If the lba + num blocks
9022	 * is less than the lba, then we've wrapped around and the block
9023	 * range is invalid anyway.
9024	 */
9025	if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9026	 || ((lba + num_blocks) < lba)) {
9027		ctl_set_lba_out_of_range(ctsio);
9028		ctl_done((union ctl_io *)ctsio);
9029		return (CTL_RETVAL_COMPLETE);
9030	}
9031
9032	/*
9033	 * According to SBC-3, a transfer length of 0 is not an error.
9034	 */
9035	if (num_blocks == 0) {
9036		ctl_set_success(ctsio);
9037		ctl_done((union ctl_io *)ctsio);
9038		return (CTL_RETVAL_COMPLETE);
9039	}
9040
9041	lbalen = (struct ctl_lba_len_flags *)
9042	    &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9043	lbalen->lba = lba;
9044	lbalen->len = num_blocks;
9045	if (bytchk) {
9046		lbalen->flags = CTL_LLF_COMPARE;
9047		ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9048	} else {
9049		lbalen->flags = CTL_LLF_VERIFY;
9050		ctsio->kern_total_len = 0;
9051	}
9052	ctsio->kern_rel_offset = 0;
9053
9054	CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9055	retval = lun->backend->data_submit((union ctl_io *)ctsio);
9056	return (retval);
9057}
9058
9059int
9060ctl_report_luns(struct ctl_scsiio *ctsio)
9061{
9062	struct scsi_report_luns *cdb;
9063	struct scsi_report_luns_data *lun_data;
9064	struct ctl_lun *lun, *request_lun;
9065	int num_luns, retval;
9066	uint32_t alloc_len, lun_datalen;
9067	int num_filled, well_known;
9068	uint32_t initidx, targ_lun_id, lun_id;
9069
9070	retval = CTL_RETVAL_COMPLETE;
9071	well_known = 0;
9072
9073	cdb = (struct scsi_report_luns *)ctsio->cdb;
9074
9075	CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9076
9077	mtx_lock(&control_softc->ctl_lock);
9078	num_luns = control_softc->num_luns;
9079	mtx_unlock(&control_softc->ctl_lock);
9080
9081	switch (cdb->select_report) {
9082	case RPL_REPORT_DEFAULT:
9083	case RPL_REPORT_ALL:
9084		break;
9085	case RPL_REPORT_WELLKNOWN:
9086		well_known = 1;
9087		num_luns = 0;
9088		break;
9089	default:
9090		ctl_set_invalid_field(ctsio,
9091				      /*sks_valid*/ 1,
9092				      /*command*/ 1,
9093				      /*field*/ 2,
9094				      /*bit_valid*/ 0,
9095				      /*bit*/ 0);
9096		ctl_done((union ctl_io *)ctsio);
9097		return (retval);
9098		break; /* NOTREACHED */
9099	}
9100
9101	alloc_len = scsi_4btoul(cdb->length);
9102	/*
9103	 * The initiator has to allocate at least 16 bytes for this request,
9104	 * so he can at least get the header and the first LUN.  Otherwise
9105	 * we reject the request (per SPC-3 rev 14, section 6.21).
9106	 */
9107	if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9108	    sizeof(struct scsi_report_luns_lundata))) {
9109		ctl_set_invalid_field(ctsio,
9110				      /*sks_valid*/ 1,
9111				      /*command*/ 1,
9112				      /*field*/ 6,
9113				      /*bit_valid*/ 0,
9114				      /*bit*/ 0);
9115		ctl_done((union ctl_io *)ctsio);
9116		return (retval);
9117	}
9118
9119	request_lun = (struct ctl_lun *)
9120		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9121
9122	lun_datalen = sizeof(*lun_data) +
9123		(num_luns * sizeof(struct scsi_report_luns_lundata));
9124
9125	ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9126	lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9127	ctsio->kern_sg_entries = 0;
9128
9129	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9130
9131	mtx_lock(&control_softc->ctl_lock);
9132	for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9133		lun_id = targ_lun_id;
9134		if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
9135			lun_id = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, lun_id);
9136		if (lun_id >= CTL_MAX_LUNS)
9137			continue;
9138		lun = control_softc->ctl_luns[lun_id];
9139		if (lun == NULL)
9140			continue;
9141
9142		if (targ_lun_id <= 0xff) {
9143			/*
9144			 * Peripheral addressing method, bus number 0.
9145			 */
9146			lun_data->luns[num_filled].lundata[0] =
9147				RPL_LUNDATA_ATYP_PERIPH;
9148			lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9149			num_filled++;
9150		} else if (targ_lun_id <= 0x3fff) {
9151			/*
9152			 * Flat addressing method.
9153			 */
9154			lun_data->luns[num_filled].lundata[0] =
9155				RPL_LUNDATA_ATYP_FLAT |
9156				(targ_lun_id & RPL_LUNDATA_FLAT_LUN_MASK);
9157#ifdef OLDCTLHEADERS
9158				(SRLD_ADDR_FLAT << SRLD_ADDR_SHIFT) |
9159				(targ_lun_id & SRLD_BUS_LUN_MASK);
9160#endif
9161			lun_data->luns[num_filled].lundata[1] =
9162#ifdef OLDCTLHEADERS
9163				targ_lun_id >> SRLD_BUS_LUN_BITS;
9164#endif
9165				targ_lun_id >> RPL_LUNDATA_FLAT_LUN_BITS;
9166			num_filled++;
9167		} else {
9168			printf("ctl_report_luns: bogus LUN number %jd, "
9169			       "skipping\n", (intmax_t)targ_lun_id);
9170		}
9171		/*
9172		 * According to SPC-3, rev 14 section 6.21:
9173		 *
9174		 * "The execution of a REPORT LUNS command to any valid and
9175		 * installed logical unit shall clear the REPORTED LUNS DATA
9176		 * HAS CHANGED unit attention condition for all logical
9177		 * units of that target with respect to the requesting
9178		 * initiator. A valid and installed logical unit is one
9179		 * having a PERIPHERAL QUALIFIER of 000b in the standard
9180		 * INQUIRY data (see 6.4.2)."
9181		 *
9182		 * If request_lun is NULL, the LUN this report luns command
9183		 * was issued to is either disabled or doesn't exist. In that
9184		 * case, we shouldn't clear any pending lun change unit
9185		 * attention.
9186		 */
9187		if (request_lun != NULL)
9188			lun->pending_sense[initidx].ua_pending &=
9189				~CTL_UA_LUN_CHANGE;
9190	}
9191	mtx_unlock(&control_softc->ctl_lock);
9192
9193	/*
9194	 * It's quite possible that we've returned fewer LUNs than we allocated
9195	 * space for.  Trim it.
9196	 */
9197	lun_datalen = sizeof(*lun_data) +
9198		(num_filled * sizeof(struct scsi_report_luns_lundata));
9199
9200	if (lun_datalen < alloc_len) {
9201		ctsio->residual = alloc_len - lun_datalen;
9202		ctsio->kern_data_len = lun_datalen;
9203		ctsio->kern_total_len = lun_datalen;
9204	} else {
9205		ctsio->residual = 0;
9206		ctsio->kern_data_len = alloc_len;
9207		ctsio->kern_total_len = alloc_len;
9208	}
9209	ctsio->kern_data_resid = 0;
9210	ctsio->kern_rel_offset = 0;
9211	ctsio->kern_sg_entries = 0;
9212
9213	/*
9214	 * We set this to the actual data length, regardless of how much
9215	 * space we actually have to return results.  If the user looks at
9216	 * this value, he'll know whether or not he allocated enough space
9217	 * and reissue the command if necessary.  We don't support well
9218	 * known logical units, so if the user asks for that, return none.
9219	 */
9220	scsi_ulto4b(lun_datalen - 8, lun_data->length);
9221
9222	/*
9223	 * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9224	 * this request.
9225	 */
9226	ctsio->scsi_status = SCSI_STATUS_OK;
9227
9228	ctsio->be_move_done = ctl_config_move_done;
9229	ctl_datamove((union ctl_io *)ctsio);
9230
9231	return (retval);
9232}
9233
9234int
9235ctl_request_sense(struct ctl_scsiio *ctsio)
9236{
9237	struct scsi_request_sense *cdb;
9238	struct scsi_sense_data *sense_ptr;
9239	struct ctl_lun *lun;
9240	uint32_t initidx;
9241	int have_error;
9242	scsi_sense_data_type sense_format;
9243
9244	cdb = (struct scsi_request_sense *)ctsio->cdb;
9245
9246	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9247
9248	CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9249
9250	/*
9251	 * Determine which sense format the user wants.
9252	 */
9253	if (cdb->byte2 & SRS_DESC)
9254		sense_format = SSD_TYPE_DESC;
9255	else
9256		sense_format = SSD_TYPE_FIXED;
9257
9258	ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9259	sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9260	ctsio->kern_sg_entries = 0;
9261
9262	/*
9263	 * struct scsi_sense_data, which is currently set to 256 bytes, is
9264	 * larger than the largest allowed value for the length field in the
9265	 * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9266	 */
9267	ctsio->residual = 0;
9268	ctsio->kern_data_len = cdb->length;
9269	ctsio->kern_total_len = cdb->length;
9270
9271	ctsio->kern_data_resid = 0;
9272	ctsio->kern_rel_offset = 0;
9273	ctsio->kern_sg_entries = 0;
9274
9275	/*
9276	 * If we don't have a LUN, we don't have any pending sense.
9277	 */
9278	if (lun == NULL)
9279		goto no_sense;
9280
9281	have_error = 0;
9282	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9283	/*
9284	 * Check for pending sense, and then for pending unit attentions.
9285	 * Pending sense gets returned first, then pending unit attentions.
9286	 */
9287	mtx_lock(&lun->ctl_softc->ctl_lock);
9288	if (ctl_is_set(lun->have_ca, initidx)) {
9289		scsi_sense_data_type stored_format;
9290
9291		/*
9292		 * Check to see which sense format was used for the stored
9293		 * sense data.
9294		 */
9295		stored_format = scsi_sense_type(
9296		    &lun->pending_sense[initidx].sense);
9297
9298		/*
9299		 * If the user requested a different sense format than the
9300		 * one we stored, then we need to convert it to the other
9301		 * format.  If we're going from descriptor to fixed format
9302		 * sense data, we may lose things in translation, depending
9303		 * on what options were used.
9304		 *
9305		 * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9306		 * for some reason we'll just copy it out as-is.
9307		 */
9308		if ((stored_format == SSD_TYPE_FIXED)
9309		 && (sense_format == SSD_TYPE_DESC))
9310			ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9311			    &lun->pending_sense[initidx].sense,
9312			    (struct scsi_sense_data_desc *)sense_ptr);
9313		else if ((stored_format == SSD_TYPE_DESC)
9314		      && (sense_format == SSD_TYPE_FIXED))
9315			ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9316			    &lun->pending_sense[initidx].sense,
9317			    (struct scsi_sense_data_fixed *)sense_ptr);
9318		else
9319			memcpy(sense_ptr, &lun->pending_sense[initidx].sense,
9320			       ctl_min(sizeof(*sense_ptr),
9321			       sizeof(lun->pending_sense[initidx].sense)));
9322
9323		ctl_clear_mask(lun->have_ca, initidx);
9324		have_error = 1;
9325	} else if (lun->pending_sense[initidx].ua_pending != CTL_UA_NONE) {
9326		ctl_ua_type ua_type;
9327
9328		ua_type = ctl_build_ua(lun->pending_sense[initidx].ua_pending,
9329				       sense_ptr, sense_format);
9330		if (ua_type != CTL_UA_NONE) {
9331			have_error = 1;
9332			/* We're reporting this UA, so clear it */
9333			lun->pending_sense[initidx].ua_pending &= ~ua_type;
9334		}
9335	}
9336	mtx_unlock(&lun->ctl_softc->ctl_lock);
9337
9338	/*
9339	 * We already have a pending error, return it.
9340	 */
9341	if (have_error != 0) {
9342		/*
9343		 * We report the SCSI status as OK, since the status of the
9344		 * request sense command itself is OK.
9345		 */
9346		ctsio->scsi_status = SCSI_STATUS_OK;
9347
9348		/*
9349		 * We report 0 for the sense length, because we aren't doing
9350		 * autosense in this case.  We're reporting sense as
9351		 * parameter data.
9352		 */
9353		ctsio->sense_len = 0;
9354
9355		ctsio->be_move_done = ctl_config_move_done;
9356		ctl_datamove((union ctl_io *)ctsio);
9357
9358		return (CTL_RETVAL_COMPLETE);
9359	}
9360
9361no_sense:
9362
9363	/*
9364	 * No sense information to report, so we report that everything is
9365	 * okay.
9366	 */
9367	ctl_set_sense_data(sense_ptr,
9368			   lun,
9369			   sense_format,
9370			   /*current_error*/ 1,
9371			   /*sense_key*/ SSD_KEY_NO_SENSE,
9372			   /*asc*/ 0x00,
9373			   /*ascq*/ 0x00,
9374			   SSD_ELEM_NONE);
9375
9376	ctsio->scsi_status = SCSI_STATUS_OK;
9377
9378	/*
9379	 * We report 0 for the sense length, because we aren't doing
9380	 * autosense in this case.  We're reporting sense as parameter data.
9381	 */
9382	ctsio->sense_len = 0;
9383	ctsio->be_move_done = ctl_config_move_done;
9384	ctl_datamove((union ctl_io *)ctsio);
9385
9386	return (CTL_RETVAL_COMPLETE);
9387}
9388
9389int
9390ctl_tur(struct ctl_scsiio *ctsio)
9391{
9392	struct ctl_lun *lun;
9393
9394	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9395
9396	CTL_DEBUG_PRINT(("ctl_tur\n"));
9397
9398	if (lun == NULL)
9399		return (EINVAL);
9400
9401	ctsio->scsi_status = SCSI_STATUS_OK;
9402	ctsio->io_hdr.status = CTL_SUCCESS;
9403
9404	ctl_done((union ctl_io *)ctsio);
9405
9406	return (CTL_RETVAL_COMPLETE);
9407}
9408
9409#ifdef notyet
9410static int
9411ctl_cmddt_inquiry(struct ctl_scsiio *ctsio)
9412{
9413
9414}
9415#endif
9416
9417static int
9418ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9419{
9420	struct scsi_vpd_supported_pages *pages;
9421	int sup_page_size;
9422	struct ctl_lun *lun;
9423
9424	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9425
9426	sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9427	    SCSI_EVPD_NUM_SUPPORTED_PAGES;
9428	ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9429	pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9430	ctsio->kern_sg_entries = 0;
9431
9432	if (sup_page_size < alloc_len) {
9433		ctsio->residual = alloc_len - sup_page_size;
9434		ctsio->kern_data_len = sup_page_size;
9435		ctsio->kern_total_len = sup_page_size;
9436	} else {
9437		ctsio->residual = 0;
9438		ctsio->kern_data_len = alloc_len;
9439		ctsio->kern_total_len = alloc_len;
9440	}
9441	ctsio->kern_data_resid = 0;
9442	ctsio->kern_rel_offset = 0;
9443	ctsio->kern_sg_entries = 0;
9444
9445	/*
9446	 * The control device is always connected.  The disk device, on the
9447	 * other hand, may not be online all the time.  Need to change this
9448	 * to figure out whether the disk device is actually online or not.
9449	 */
9450	if (lun != NULL)
9451		pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9452				lun->be_lun->lun_type;
9453	else
9454		pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9455
9456	pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES;
9457	/* Supported VPD pages */
9458	pages->page_list[0] = SVPD_SUPPORTED_PAGES;
9459	/* Serial Number */
9460	pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER;
9461	/* Device Identification */
9462	pages->page_list[2] = SVPD_DEVICE_ID;
9463	/* Block limits */
9464	pages->page_list[3] = SVPD_BLOCK_LIMITS;
9465	/* Logical Block Provisioning */
9466	pages->page_list[4] = SVPD_LBP;
9467
9468	ctsio->scsi_status = SCSI_STATUS_OK;
9469
9470	ctsio->be_move_done = ctl_config_move_done;
9471	ctl_datamove((union ctl_io *)ctsio);
9472
9473	return (CTL_RETVAL_COMPLETE);
9474}
9475
9476static int
9477ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9478{
9479	struct scsi_vpd_unit_serial_number *sn_ptr;
9480	struct ctl_lun *lun;
9481#ifndef CTL_USE_BACKEND_SN
9482	char tmpstr[32];
9483#endif
9484
9485	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9486
9487	ctsio->kern_data_ptr = malloc(sizeof(*sn_ptr), M_CTL, M_WAITOK | M_ZERO);
9488	sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9489	ctsio->kern_sg_entries = 0;
9490
9491	if (sizeof(*sn_ptr) < alloc_len) {
9492		ctsio->residual = alloc_len - sizeof(*sn_ptr);
9493		ctsio->kern_data_len = sizeof(*sn_ptr);
9494		ctsio->kern_total_len = sizeof(*sn_ptr);
9495	} else {
9496		ctsio->residual = 0;
9497		ctsio->kern_data_len = alloc_len;
9498		ctsio->kern_total_len = alloc_len;
9499	}
9500	ctsio->kern_data_resid = 0;
9501	ctsio->kern_rel_offset = 0;
9502	ctsio->kern_sg_entries = 0;
9503
9504	/*
9505	 * The control device is always connected.  The disk device, on the
9506	 * other hand, may not be online all the time.  Need to change this
9507	 * to figure out whether the disk device is actually online or not.
9508	 */
9509	if (lun != NULL)
9510		sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9511				  lun->be_lun->lun_type;
9512	else
9513		sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9514
9515	sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9516	sn_ptr->length = ctl_min(sizeof(*sn_ptr) - 4, CTL_SN_LEN);
9517#ifdef CTL_USE_BACKEND_SN
9518	/*
9519	 * If we don't have a LUN, we just leave the serial number as
9520	 * all spaces.
9521	 */
9522	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9523	if (lun != NULL) {
9524		strncpy((char *)sn_ptr->serial_num,
9525			(char *)lun->be_lun->serial_num, CTL_SN_LEN);
9526	}
9527#else
9528	/*
9529	 * Note that we're using a non-unique serial number here,
9530	 */
9531	snprintf(tmpstr, sizeof(tmpstr), "MYSERIALNUMIS000");
9532	memset(sn_ptr->serial_num, 0x20, sizeof(sn_ptr->serial_num));
9533	strncpy(sn_ptr->serial_num, tmpstr, ctl_min(CTL_SN_LEN,
9534		ctl_min(sizeof(tmpstr), sizeof(*sn_ptr) - 4)));
9535#endif
9536	ctsio->scsi_status = SCSI_STATUS_OK;
9537
9538	ctsio->be_move_done = ctl_config_move_done;
9539	ctl_datamove((union ctl_io *)ctsio);
9540
9541	return (CTL_RETVAL_COMPLETE);
9542}
9543
9544
9545static int
9546ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9547{
9548	struct scsi_vpd_device_id *devid_ptr;
9549	struct scsi_vpd_id_descriptor *desc, *desc1;
9550	struct scsi_vpd_id_descriptor *desc2, *desc3; /* for types 4h and 5h */
9551	struct scsi_vpd_id_t10 *t10id;
9552	struct ctl_softc *ctl_softc;
9553	struct ctl_lun *lun;
9554	struct ctl_frontend *fe;
9555	char *val;
9556#ifndef CTL_USE_BACKEND_SN
9557	char tmpstr[32];
9558#endif /* CTL_USE_BACKEND_SN */
9559	int devid_len;
9560
9561	ctl_softc = control_softc;
9562
9563	mtx_lock(&ctl_softc->ctl_lock);
9564	fe = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)];
9565	mtx_unlock(&ctl_softc->ctl_lock);
9566
9567	if (fe->devid != NULL)
9568		return ((fe->devid)(ctsio, alloc_len));
9569
9570	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9571
9572	devid_len = sizeof(struct scsi_vpd_device_id) +
9573		sizeof(struct scsi_vpd_id_descriptor) +
9574		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN +
9575		sizeof(struct scsi_vpd_id_descriptor) + CTL_WWPN_LEN +
9576		sizeof(struct scsi_vpd_id_descriptor) +
9577		sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9578		sizeof(struct scsi_vpd_id_descriptor) +
9579		sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9580
9581	ctsio->kern_data_ptr = malloc(devid_len, M_CTL, M_WAITOK | M_ZERO);
9582	devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9583	ctsio->kern_sg_entries = 0;
9584
9585	if (devid_len < alloc_len) {
9586		ctsio->residual = alloc_len - devid_len;
9587		ctsio->kern_data_len = devid_len;
9588		ctsio->kern_total_len = devid_len;
9589	} else {
9590		ctsio->residual = 0;
9591		ctsio->kern_data_len = alloc_len;
9592		ctsio->kern_total_len = alloc_len;
9593	}
9594	ctsio->kern_data_resid = 0;
9595	ctsio->kern_rel_offset = 0;
9596	ctsio->kern_sg_entries = 0;
9597
9598	desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9599	t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
9600	desc1 = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9601		sizeof(struct scsi_vpd_id_t10) + CTL_DEVID_LEN);
9602	desc2 = (struct scsi_vpd_id_descriptor *)(&desc1->identifier[0] +
9603	          CTL_WWPN_LEN);
9604	desc3 = (struct scsi_vpd_id_descriptor *)(&desc2->identifier[0] +
9605	         sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9606
9607	/*
9608	 * The control device is always connected.  The disk device, on the
9609	 * other hand, may not be online all the time.
9610	 */
9611	if (lun != NULL)
9612		devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9613				     lun->be_lun->lun_type;
9614	else
9615		devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9616
9617	devid_ptr->page_code = SVPD_DEVICE_ID;
9618
9619	scsi_ulto2b(devid_len - 4, devid_ptr->length);
9620
9621	mtx_lock(&ctl_softc->ctl_lock);
9622
9623	/*
9624	 * For Fibre channel,
9625	 */
9626	if (fe->port_type == CTL_PORT_FC)
9627	{
9628		desc->proto_codeset = (SCSI_PROTO_FC << 4) |
9629				      SVPD_ID_CODESET_ASCII;
9630        	desc1->proto_codeset = (SCSI_PROTO_FC << 4) |
9631		              SVPD_ID_CODESET_BINARY;
9632	}
9633	else
9634	{
9635		desc->proto_codeset = (SCSI_PROTO_SPI << 4) |
9636				      SVPD_ID_CODESET_ASCII;
9637        	desc1->proto_codeset = (SCSI_PROTO_SPI << 4) |
9638		              SVPD_ID_CODESET_BINARY;
9639	}
9640	desc2->proto_codeset = desc3->proto_codeset = desc1->proto_codeset;
9641	mtx_unlock(&ctl_softc->ctl_lock);
9642
9643	/*
9644	 * We're using a LUN association here.  i.e., this device ID is a
9645	 * per-LUN identifier.
9646	 */
9647	desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
9648	desc->length = sizeof(*t10id) + CTL_DEVID_LEN;
9649	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
9650		strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
9651	} else {
9652		memset(t10id->vendor, ' ', sizeof(t10id->vendor));
9653		strncpy(t10id->vendor, val,
9654		    min(sizeof(t10id->vendor), strlen(val)));
9655	}
9656
9657	/*
9658	 * desc1 is for the WWPN which is a port asscociation.
9659	 */
9660	desc1->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA;
9661	desc1->length = CTL_WWPN_LEN;
9662	/* XXX Call Reggie's get_WWNN func here then add port # to the end */
9663	/* For testing just create the WWPN */
9664#if 0
9665	ddb_GetWWNN((char *)desc1->identifier);
9666
9667	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9668	/* This is so Copancontrol will return something sane */
9669	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9670	    ctsio->io_hdr.nexus.targ_port!=8)
9671		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port-1;
9672	else
9673		desc1->identifier[7] += ctsio->io_hdr.nexus.targ_port;
9674#endif
9675
9676	be64enc(desc1->identifier, fe->wwpn);
9677
9678	/*
9679	 * desc2 is for the Relative Target Port(type 4h) identifier
9680	 */
9681	desc2->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9682	                 | SVPD_ID_TYPE_RELTARG;
9683	desc2->length = 4;
9684//#if 0
9685	/* NOTE: if the port is 0 or 8 we don't want to subtract 1 */
9686	/* This is so Copancontrol will return something sane */
9687	if (ctsio->io_hdr.nexus.targ_port!=0 &&
9688	    ctsio->io_hdr.nexus.targ_port!=8)
9689		desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port - 1;
9690	else
9691	        desc2->identifier[3] = ctsio->io_hdr.nexus.targ_port;
9692//#endif
9693
9694	/*
9695	 * desc3 is for the Target Port Group(type 5h) identifier
9696	 */
9697	desc3->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT
9698	                 | SVPD_ID_TYPE_TPORTGRP;
9699	desc3->length = 4;
9700	if (ctsio->io_hdr.nexus.targ_port < CTL_MAX_PORTS || ctl_is_single)
9701		desc3->identifier[3] = 1;
9702	else
9703		desc3->identifier[3] = 2;
9704
9705#ifdef CTL_USE_BACKEND_SN
9706	/*
9707	 * If we've actually got a backend, copy the device id from the
9708	 * per-LUN data.  Otherwise, set it to all spaces.
9709	 */
9710	if (lun != NULL) {
9711		/*
9712		 * Copy the backend's LUN ID.
9713		 */
9714		strncpy((char *)t10id->vendor_spec_id,
9715			(char *)lun->be_lun->device_id, CTL_DEVID_LEN);
9716	} else {
9717		/*
9718		 * No backend, set this to spaces.
9719		 */
9720		memset(t10id->vendor_spec_id, 0x20, CTL_DEVID_LEN);
9721	}
9722#else
9723	snprintf(tmpstr, sizeof(tmpstr), "MYDEVICEIDIS%4d",
9724		 (lun != NULL) ?  (int)lun->lun : 0);
9725	strncpy(t10id->vendor_spec_id, tmpstr, ctl_min(CTL_DEVID_LEN,
9726		sizeof(tmpstr)));
9727#endif
9728
9729	ctsio->scsi_status = SCSI_STATUS_OK;
9730
9731	ctsio->be_move_done = ctl_config_move_done;
9732	ctl_datamove((union ctl_io *)ctsio);
9733
9734	return (CTL_RETVAL_COMPLETE);
9735}
9736
9737static int
9738ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9739{
9740	struct scsi_vpd_block_limits *bl_ptr;
9741	struct ctl_lun *lun;
9742	int bs;
9743
9744	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9745	bs = lun->be_lun->blocksize;
9746
9747	ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9748	bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9749	ctsio->kern_sg_entries = 0;
9750
9751	if (sizeof(*bl_ptr) < alloc_len) {
9752		ctsio->residual = alloc_len - sizeof(*bl_ptr);
9753		ctsio->kern_data_len = sizeof(*bl_ptr);
9754		ctsio->kern_total_len = sizeof(*bl_ptr);
9755	} else {
9756		ctsio->residual = 0;
9757		ctsio->kern_data_len = alloc_len;
9758		ctsio->kern_total_len = alloc_len;
9759	}
9760	ctsio->kern_data_resid = 0;
9761	ctsio->kern_rel_offset = 0;
9762	ctsio->kern_sg_entries = 0;
9763
9764	/*
9765	 * The control device is always connected.  The disk device, on the
9766	 * other hand, may not be online all the time.  Need to change this
9767	 * to figure out whether the disk device is actually online or not.
9768	 */
9769	if (lun != NULL)
9770		bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9771				  lun->be_lun->lun_type;
9772	else
9773		bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9774
9775	bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9776	scsi_ulto2b(sizeof(*bl_ptr), bl_ptr->page_length);
9777	bl_ptr->max_cmp_write_len = 0xff;
9778	scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9779	scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len);
9780	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9781		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9782		scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9783	}
9784	scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9785
9786	ctsio->scsi_status = SCSI_STATUS_OK;
9787	ctsio->be_move_done = ctl_config_move_done;
9788	ctl_datamove((union ctl_io *)ctsio);
9789
9790	return (CTL_RETVAL_COMPLETE);
9791}
9792
9793static int
9794ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9795{
9796	struct scsi_vpd_logical_block_prov *lbp_ptr;
9797	struct ctl_lun *lun;
9798	int bs;
9799
9800	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9801	bs = lun->be_lun->blocksize;
9802
9803	ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9804	lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9805	ctsio->kern_sg_entries = 0;
9806
9807	if (sizeof(*lbp_ptr) < alloc_len) {
9808		ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9809		ctsio->kern_data_len = sizeof(*lbp_ptr);
9810		ctsio->kern_total_len = sizeof(*lbp_ptr);
9811	} else {
9812		ctsio->residual = 0;
9813		ctsio->kern_data_len = alloc_len;
9814		ctsio->kern_total_len = alloc_len;
9815	}
9816	ctsio->kern_data_resid = 0;
9817	ctsio->kern_rel_offset = 0;
9818	ctsio->kern_sg_entries = 0;
9819
9820	/*
9821	 * The control device is always connected.  The disk device, on the
9822	 * other hand, may not be online all the time.  Need to change this
9823	 * to figure out whether the disk device is actually online or not.
9824	 */
9825	if (lun != NULL)
9826		lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9827				  lun->be_lun->lun_type;
9828	else
9829		lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9830
9831	lbp_ptr->page_code = SVPD_LBP;
9832	if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
9833		lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10;
9834
9835	ctsio->scsi_status = SCSI_STATUS_OK;
9836	ctsio->be_move_done = ctl_config_move_done;
9837	ctl_datamove((union ctl_io *)ctsio);
9838
9839	return (CTL_RETVAL_COMPLETE);
9840}
9841
9842static int
9843ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9844{
9845	struct scsi_inquiry *cdb;
9846	struct ctl_lun *lun;
9847	int alloc_len, retval;
9848
9849	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9850	cdb = (struct scsi_inquiry *)ctsio->cdb;
9851
9852	retval = CTL_RETVAL_COMPLETE;
9853
9854	alloc_len = scsi_2btoul(cdb->length);
9855
9856	switch (cdb->page_code) {
9857	case SVPD_SUPPORTED_PAGES:
9858		retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9859		break;
9860	case SVPD_UNIT_SERIAL_NUMBER:
9861		retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9862		break;
9863	case SVPD_DEVICE_ID:
9864		retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9865		break;
9866	case SVPD_BLOCK_LIMITS:
9867		retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9868		break;
9869	case SVPD_LBP:
9870		retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9871		break;
9872	default:
9873		ctl_set_invalid_field(ctsio,
9874				      /*sks_valid*/ 1,
9875				      /*command*/ 1,
9876				      /*field*/ 2,
9877				      /*bit_valid*/ 0,
9878				      /*bit*/ 0);
9879		ctl_done((union ctl_io *)ctsio);
9880		retval = CTL_RETVAL_COMPLETE;
9881		break;
9882	}
9883
9884	return (retval);
9885}
9886
9887static int
9888ctl_inquiry_std(struct ctl_scsiio *ctsio)
9889{
9890	struct scsi_inquiry_data *inq_ptr;
9891	struct scsi_inquiry *cdb;
9892	struct ctl_softc *ctl_softc;
9893	struct ctl_lun *lun;
9894	char *val;
9895	uint32_t alloc_len;
9896	int is_fc;
9897
9898	ctl_softc = control_softc;
9899
9900	/*
9901	 * Figure out whether we're talking to a Fibre Channel port or not.
9902	 * We treat the ioctl front end, and any SCSI adapters, as packetized
9903	 * SCSI front ends.
9904	 */
9905	mtx_lock(&ctl_softc->ctl_lock);
9906	if (ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type !=
9907	    CTL_PORT_FC)
9908		is_fc = 0;
9909	else
9910		is_fc = 1;
9911	mtx_unlock(&ctl_softc->ctl_lock);
9912
9913	lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9914	cdb = (struct scsi_inquiry *)ctsio->cdb;
9915	alloc_len = scsi_2btoul(cdb->length);
9916
9917	/*
9918	 * We malloc the full inquiry data size here and fill it
9919	 * in.  If the user only asks for less, we'll give him
9920	 * that much.
9921	 */
9922	ctsio->kern_data_ptr = malloc(sizeof(*inq_ptr), M_CTL, M_WAITOK | M_ZERO);
9923	inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9924	ctsio->kern_sg_entries = 0;
9925	ctsio->kern_data_resid = 0;
9926	ctsio->kern_rel_offset = 0;
9927
9928	if (sizeof(*inq_ptr) < alloc_len) {
9929		ctsio->residual = alloc_len - sizeof(*inq_ptr);
9930		ctsio->kern_data_len = sizeof(*inq_ptr);
9931		ctsio->kern_total_len = sizeof(*inq_ptr);
9932	} else {
9933		ctsio->residual = 0;
9934		ctsio->kern_data_len = alloc_len;
9935		ctsio->kern_total_len = alloc_len;
9936	}
9937
9938	/*
9939	 * If we have a LUN configured, report it as connected.  Otherwise,
9940	 * report that it is offline or no device is supported, depending
9941	 * on the value of inquiry_pq_no_lun.
9942	 *
9943	 * According to the spec (SPC-4 r34), the peripheral qualifier
9944	 * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario:
9945	 *
9946	 * "A peripheral device having the specified peripheral device type
9947	 * is not connected to this logical unit. However, the device
9948	 * server is capable of supporting the specified peripheral device
9949	 * type on this logical unit."
9950	 *
9951	 * According to the same spec, the peripheral qualifier
9952	 * SID_QUAL_BAD_LU (011b) is used in this scenario:
9953	 *
9954	 * "The device server is not capable of supporting a peripheral
9955	 * device on this logical unit. For this peripheral qualifier the
9956	 * peripheral device type shall be set to 1Fh. All other peripheral
9957	 * device type values are reserved for this peripheral qualifier."
9958	 *
9959	 * Given the text, it would seem that we probably want to report that
9960	 * the LUN is offline here.  There is no LUN connected, but we can
9961	 * support a LUN at the given LUN number.
9962	 *
9963	 * In the real world, though, it sounds like things are a little
9964	 * different:
9965	 *
9966	 * - Linux, when presented with a LUN with the offline peripheral
9967	 *   qualifier, will create an sg driver instance for it.  So when
9968	 *   you attach it to CTL, you wind up with a ton of sg driver
9969	 *   instances.  (One for every LUN that Linux bothered to probe.)
9970	 *   Linux does this despite the fact that it issues a REPORT LUNs
9971	 *   to LUN 0 to get the inventory of supported LUNs.
9972	 *
9973	 * - There is other anecdotal evidence (from Emulex folks) about
9974	 *   arrays that use the offline peripheral qualifier for LUNs that
9975	 *   are on the "passive" path in an active/passive array.
9976	 *
9977	 * So the solution is provide a hopefully reasonable default
9978	 * (return bad/no LUN) and allow the user to change the behavior
9979	 * with a tunable/sysctl variable.
9980	 */
9981	if (lun != NULL)
9982		inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9983				  lun->be_lun->lun_type;
9984	else if (ctl_softc->inquiry_pq_no_lun == 0)
9985		inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9986	else
9987		inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9988
9989	/* RMB in byte 2 is 0 */
9990	inq_ptr->version = SCSI_REV_SPC3;
9991
9992	/*
9993	 * According to SAM-3, even if a device only supports a single
9994	 * level of LUN addressing, it should still set the HISUP bit:
9995	 *
9996	 * 4.9.1 Logical unit numbers overview
9997	 *
9998	 * All logical unit number formats described in this standard are
9999	 * hierarchical in structure even when only a single level in that
10000	 * hierarchy is used. The HISUP bit shall be set to one in the
10001	 * standard INQUIRY data (see SPC-2) when any logical unit number
10002	 * format described in this standard is used.  Non-hierarchical
10003	 * formats are outside the scope of this standard.
10004	 *
10005	 * Therefore we set the HiSup bit here.
10006	 *
10007	 * The reponse format is 2, per SPC-3.
10008	 */
10009	inq_ptr->response_format = SID_HiSup | 2;
10010
10011	inq_ptr->additional_length = sizeof(*inq_ptr) - 4;
10012	CTL_DEBUG_PRINT(("additional_length = %d\n",
10013			 inq_ptr->additional_length));
10014
10015	inq_ptr->spc3_flags = SPC3_SID_TPGS_IMPLICIT;
10016	/* 16 bit addressing */
10017	if (is_fc == 0)
10018		inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10019	/* XXX set the SID_MultiP bit here if we're actually going to
10020	   respond on multiple ports */
10021	inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10022
10023	/* 16 bit data bus, synchronous transfers */
10024	/* XXX these flags don't apply for FC */
10025	if (is_fc == 0)
10026		inq_ptr->flags = SID_WBus16 | SID_Sync;
10027	/*
10028	 * XXX KDM do we want to support tagged queueing on the control
10029	 * device at all?
10030	 */
10031	if ((lun == NULL)
10032	 || (lun->be_lun->lun_type != T_PROCESSOR))
10033		inq_ptr->flags |= SID_CmdQue;
10034	/*
10035	 * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10036	 * We have 8 bytes for the vendor name, and 16 bytes for the device
10037	 * name and 4 bytes for the revision.
10038	 */
10039	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "vendor")) == NULL) {
10040		strcpy(inq_ptr->vendor, CTL_VENDOR);
10041	} else {
10042		memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10043		strncpy(inq_ptr->vendor, val,
10044		    min(sizeof(inq_ptr->vendor), strlen(val)));
10045	}
10046	if (lun == NULL) {
10047		strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10048	} else if ((val = ctl_get_opt(lun->be_lun, "product")) == NULL) {
10049		switch (lun->be_lun->lun_type) {
10050		case T_DIRECT:
10051			strcpy(inq_ptr->product, CTL_DIRECT_PRODUCT);
10052			break;
10053		case T_PROCESSOR:
10054			strcpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT);
10055			break;
10056		default:
10057			strcpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT);
10058			break;
10059		}
10060	} else {
10061		memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10062		strncpy(inq_ptr->product, val,
10063		    min(sizeof(inq_ptr->product), strlen(val)));
10064	}
10065
10066	/*
10067	 * XXX make this a macro somewhere so it automatically gets
10068	 * incremented when we make changes.
10069	 */
10070	if (lun == NULL || (val = ctl_get_opt(lun->be_lun, "revision")) == NULL) {
10071		strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10072	} else {
10073		memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10074		strncpy(inq_ptr->revision, val,
10075		    min(sizeof(inq_ptr->revision), strlen(val)));
10076	}
10077
10078	/*
10079	 * For parallel SCSI, we support double transition and single
10080	 * transition clocking.  We also support QAS (Quick Arbitration
10081	 * and Selection) and Information Unit transfers on both the
10082	 * control and array devices.
10083	 */
10084	if (is_fc == 0)
10085		inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10086				    SID_SPI_IUS;
10087
10088	/* SAM-3 */
10089	scsi_ulto2b(0x0060, inq_ptr->version1);
10090	/* SPC-3 (no version claimed) XXX should we claim a version? */
10091	scsi_ulto2b(0x0300, inq_ptr->version2);
10092	if (is_fc) {
10093		/* FCP-2 ANSI INCITS.350:2003 */
10094		scsi_ulto2b(0x0917, inq_ptr->version3);
10095	} else {
10096		/* SPI-4 ANSI INCITS.362:200x */
10097		scsi_ulto2b(0x0B56, inq_ptr->version3);
10098	}
10099
10100	if (lun == NULL) {
10101		/* SBC-2 (no version claimed) XXX should we claim a version? */
10102		scsi_ulto2b(0x0320, inq_ptr->version4);
10103	} else {
10104		switch (lun->be_lun->lun_type) {
10105		case T_DIRECT:
10106			/*
10107			 * SBC-2 (no version claimed) XXX should we claim a
10108			 * version?
10109			 */
10110			scsi_ulto2b(0x0320, inq_ptr->version4);
10111			break;
10112		case T_PROCESSOR:
10113		default:
10114			break;
10115		}
10116	}
10117
10118	ctsio->scsi_status = SCSI_STATUS_OK;
10119	if (ctsio->kern_data_len > 0) {
10120		ctsio->be_move_done = ctl_config_move_done;
10121		ctl_datamove((union ctl_io *)ctsio);
10122	} else {
10123		ctsio->io_hdr.status = CTL_SUCCESS;
10124		ctl_done((union ctl_io *)ctsio);
10125	}
10126
10127	return (CTL_RETVAL_COMPLETE);
10128}
10129
10130int
10131ctl_inquiry(struct ctl_scsiio *ctsio)
10132{
10133	struct scsi_inquiry *cdb;
10134	int retval;
10135
10136	cdb = (struct scsi_inquiry *)ctsio->cdb;
10137
10138	retval = 0;
10139
10140	CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10141
10142	/*
10143	 * Right now, we don't support the CmdDt inquiry information.
10144	 * This would be nice to support in the future.  When we do
10145	 * support it, we should change this test so that it checks to make
10146	 * sure SI_EVPD and SI_CMDDT aren't both set at the same time.
10147	 */
10148#ifdef notyet
10149	if (((cdb->byte2 & SI_EVPD)
10150	 && (cdb->byte2 & SI_CMDDT)))
10151#endif
10152	if (cdb->byte2 & SI_CMDDT) {
10153		/*
10154		 * Point to the SI_CMDDT bit.  We might change this
10155		 * when we support SI_CMDDT, but since both bits would be
10156		 * "wrong", this should probably just stay as-is then.
10157		 */
10158		ctl_set_invalid_field(ctsio,
10159				      /*sks_valid*/ 1,
10160				      /*command*/ 1,
10161				      /*field*/ 1,
10162				      /*bit_valid*/ 1,
10163				      /*bit*/ 1);
10164		ctl_done((union ctl_io *)ctsio);
10165		return (CTL_RETVAL_COMPLETE);
10166	}
10167	if (cdb->byte2 & SI_EVPD)
10168		retval = ctl_inquiry_evpd(ctsio);
10169#ifdef notyet
10170	else if (cdb->byte2 & SI_CMDDT)
10171		retval = ctl_inquiry_cmddt(ctsio);
10172#endif
10173	else
10174		retval = ctl_inquiry_std(ctsio);
10175
10176	return (retval);
10177}
10178
10179/*
10180 * For known CDB types, parse the LBA and length.
10181 */
10182static int
10183ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint32_t *len)
10184{
10185	if (io->io_hdr.io_type != CTL_IO_SCSI)
10186		return (1);
10187
10188	switch (io->scsiio.cdb[0]) {
10189	case COMPARE_AND_WRITE: {
10190		struct scsi_compare_and_write *cdb;
10191
10192		cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10193
10194		*lba = scsi_8btou64(cdb->addr);
10195		*len = cdb->length;
10196		break;
10197	}
10198	case READ_6:
10199	case WRITE_6: {
10200		struct scsi_rw_6 *cdb;
10201
10202		cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10203
10204		*lba = scsi_3btoul(cdb->addr);
10205		/* only 5 bits are valid in the most significant address byte */
10206		*lba &= 0x1fffff;
10207		*len = cdb->length;
10208		break;
10209	}
10210	case READ_10:
10211	case WRITE_10: {
10212		struct scsi_rw_10 *cdb;
10213
10214		cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10215
10216		*lba = scsi_4btoul(cdb->addr);
10217		*len = scsi_2btoul(cdb->length);
10218		break;
10219	}
10220	case WRITE_VERIFY_10: {
10221		struct scsi_write_verify_10 *cdb;
10222
10223		cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10224
10225		*lba = scsi_4btoul(cdb->addr);
10226		*len = scsi_2btoul(cdb->length);
10227		break;
10228	}
10229	case READ_12:
10230	case WRITE_12: {
10231		struct scsi_rw_12 *cdb;
10232
10233		cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10234
10235		*lba = scsi_4btoul(cdb->addr);
10236		*len = scsi_4btoul(cdb->length);
10237		break;
10238	}
10239	case WRITE_VERIFY_12: {
10240		struct scsi_write_verify_12 *cdb;
10241
10242		cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10243
10244		*lba = scsi_4btoul(cdb->addr);
10245		*len = scsi_4btoul(cdb->length);
10246		break;
10247	}
10248	case READ_16:
10249	case WRITE_16: {
10250		struct scsi_rw_16 *cdb;
10251
10252		cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10253
10254		*lba = scsi_8btou64(cdb->addr);
10255		*len = scsi_4btoul(cdb->length);
10256		break;
10257	}
10258	case WRITE_VERIFY_16: {
10259		struct scsi_write_verify_16 *cdb;
10260
10261		cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10262
10263
10264		*lba = scsi_8btou64(cdb->addr);
10265		*len = scsi_4btoul(cdb->length);
10266		break;
10267	}
10268	case WRITE_SAME_10: {
10269		struct scsi_write_same_10 *cdb;
10270
10271		cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10272
10273		*lba = scsi_4btoul(cdb->addr);
10274		*len = scsi_2btoul(cdb->length);
10275		break;
10276	}
10277	case WRITE_SAME_16: {
10278		struct scsi_write_same_16 *cdb;
10279
10280		cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10281
10282		*lba = scsi_8btou64(cdb->addr);
10283		*len = scsi_4btoul(cdb->length);
10284		break;
10285	}
10286	case VERIFY_10: {
10287		struct scsi_verify_10 *cdb;
10288
10289		cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10290
10291		*lba = scsi_4btoul(cdb->addr);
10292		*len = scsi_2btoul(cdb->length);
10293		break;
10294	}
10295	case VERIFY_12: {
10296		struct scsi_verify_12 *cdb;
10297
10298		cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10299
10300		*lba = scsi_4btoul(cdb->addr);
10301		*len = scsi_4btoul(cdb->length);
10302		break;
10303	}
10304	case VERIFY_16: {
10305		struct scsi_verify_16 *cdb;
10306
10307		cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10308
10309		*lba = scsi_8btou64(cdb->addr);
10310		*len = scsi_4btoul(cdb->length);
10311		break;
10312	}
10313	default:
10314		return (1);
10315		break; /* NOTREACHED */
10316	}
10317
10318	return (0);
10319}
10320
10321static ctl_action
10322ctl_extent_check_lba(uint64_t lba1, uint32_t len1, uint64_t lba2, uint32_t len2)
10323{
10324	uint64_t endlba1, endlba2;
10325
10326	endlba1 = lba1 + len1 - 1;
10327	endlba2 = lba2 + len2 - 1;
10328
10329	if ((endlba1 < lba2)
10330	 || (endlba2 < lba1))
10331		return (CTL_ACTION_PASS);
10332	else
10333		return (CTL_ACTION_BLOCK);
10334}
10335
10336static ctl_action
10337ctl_extent_check(union ctl_io *io1, union ctl_io *io2)
10338{
10339	uint64_t lba1, lba2;
10340	uint32_t len1, len2;
10341	int retval;
10342
10343	retval = ctl_get_lba_len(io1, &lba1, &len1);
10344	if (retval != 0)
10345		return (CTL_ACTION_ERROR);
10346
10347	retval = ctl_get_lba_len(io2, &lba2, &len2);
10348	if (retval != 0)
10349		return (CTL_ACTION_ERROR);
10350
10351	return (ctl_extent_check_lba(lba1, len1, lba2, len2));
10352}
10353
10354static ctl_action
10355ctl_check_for_blockage(union ctl_io *pending_io, union ctl_io *ooa_io)
10356{
10357	struct ctl_cmd_entry *pending_entry, *ooa_entry;
10358	ctl_serialize_action *serialize_row;
10359
10360	/*
10361	 * The initiator attempted multiple untagged commands at the same
10362	 * time.  Can't do that.
10363	 */
10364	if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10365	 && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10366	 && ((pending_io->io_hdr.nexus.targ_port ==
10367	      ooa_io->io_hdr.nexus.targ_port)
10368	  && (pending_io->io_hdr.nexus.initid.id ==
10369	      ooa_io->io_hdr.nexus.initid.id))
10370	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10371		return (CTL_ACTION_OVERLAP);
10372
10373	/*
10374	 * The initiator attempted to send multiple tagged commands with
10375	 * the same ID.  (It's fine if different initiators have the same
10376	 * tag ID.)
10377	 *
10378	 * Even if all of those conditions are true, we don't kill the I/O
10379	 * if the command ahead of us has been aborted.  We won't end up
10380	 * sending it to the FETD, and it's perfectly legal to resend a
10381	 * command with the same tag number as long as the previous
10382	 * instance of this tag number has been aborted somehow.
10383	 */
10384	if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10385	 && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10386	 && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10387	 && ((pending_io->io_hdr.nexus.targ_port ==
10388	      ooa_io->io_hdr.nexus.targ_port)
10389	  && (pending_io->io_hdr.nexus.initid.id ==
10390	      ooa_io->io_hdr.nexus.initid.id))
10391	 && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 0))
10392		return (CTL_ACTION_OVERLAP_TAG);
10393
10394	/*
10395	 * If we get a head of queue tag, SAM-3 says that we should
10396	 * immediately execute it.
10397	 *
10398	 * What happens if this command would normally block for some other
10399	 * reason?  e.g. a request sense with a head of queue tag
10400	 * immediately after a write.  Normally that would block, but this
10401	 * will result in its getting executed immediately...
10402	 *
10403	 * We currently return "pass" instead of "skip", so we'll end up
10404	 * going through the rest of the queue to check for overlapped tags.
10405	 *
10406	 * XXX KDM check for other types of blockage first??
10407	 */
10408	if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10409		return (CTL_ACTION_PASS);
10410
10411	/*
10412	 * Ordered tags have to block until all items ahead of them
10413	 * have completed.  If we get called with an ordered tag, we always
10414	 * block, if something else is ahead of us in the queue.
10415	 */
10416	if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10417		return (CTL_ACTION_BLOCK);
10418
10419	/*
10420	 * Simple tags get blocked until all head of queue and ordered tags
10421	 * ahead of them have completed.  I'm lumping untagged commands in
10422	 * with simple tags here.  XXX KDM is that the right thing to do?
10423	 */
10424	if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10425	  || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10426	 && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10427	  || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10428		return (CTL_ACTION_BLOCK);
10429
10430	pending_entry = &ctl_cmd_table[pending_io->scsiio.cdb[0]];
10431	ooa_entry = &ctl_cmd_table[ooa_io->scsiio.cdb[0]];
10432
10433	serialize_row = ctl_serialize_table[ooa_entry->seridx];
10434
10435	switch (serialize_row[pending_entry->seridx]) {
10436	case CTL_SER_BLOCK:
10437		return (CTL_ACTION_BLOCK);
10438		break; /* NOTREACHED */
10439	case CTL_SER_EXTENT:
10440		return (ctl_extent_check(pending_io, ooa_io));
10441		break; /* NOTREACHED */
10442	case CTL_SER_PASS:
10443		return (CTL_ACTION_PASS);
10444		break; /* NOTREACHED */
10445	case CTL_SER_SKIP:
10446		return (CTL_ACTION_SKIP);
10447		break;
10448	default:
10449		panic("invalid serialization value %d",
10450		      serialize_row[pending_entry->seridx]);
10451		break; /* NOTREACHED */
10452	}
10453
10454	return (CTL_ACTION_ERROR);
10455}
10456
10457/*
10458 * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10459 * Assumptions:
10460 * - pending_io is generally either incoming, or on the blocked queue
10461 * - starting I/O is the I/O we want to start the check with.
10462 */
10463static ctl_action
10464ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10465	      union ctl_io *starting_io)
10466{
10467	union ctl_io *ooa_io;
10468	ctl_action action;
10469
10470	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
10471
10472	/*
10473	 * Run back along the OOA queue, starting with the current
10474	 * blocked I/O and going through every I/O before it on the
10475	 * queue.  If starting_io is NULL, we'll just end up returning
10476	 * CTL_ACTION_PASS.
10477	 */
10478	for (ooa_io = starting_io; ooa_io != NULL;
10479	     ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10480	     ooa_links)){
10481
10482		/*
10483		 * This routine just checks to see whether
10484		 * cur_blocked is blocked by ooa_io, which is ahead
10485		 * of it in the queue.  It doesn't queue/dequeue
10486		 * cur_blocked.
10487		 */
10488		action = ctl_check_for_blockage(pending_io, ooa_io);
10489		switch (action) {
10490		case CTL_ACTION_BLOCK:
10491		case CTL_ACTION_OVERLAP:
10492		case CTL_ACTION_OVERLAP_TAG:
10493		case CTL_ACTION_SKIP:
10494		case CTL_ACTION_ERROR:
10495			return (action);
10496			break; /* NOTREACHED */
10497		case CTL_ACTION_PASS:
10498			break;
10499		default:
10500			panic("invalid action %d", action);
10501			break;  /* NOTREACHED */
10502		}
10503	}
10504
10505	return (CTL_ACTION_PASS);
10506}
10507
10508/*
10509 * Assumptions:
10510 * - An I/O has just completed, and has been removed from the per-LUN OOA
10511 *   queue, so some items on the blocked queue may now be unblocked.
10512 */
10513static int
10514ctl_check_blocked(struct ctl_lun *lun)
10515{
10516	union ctl_io *cur_blocked, *next_blocked;
10517
10518	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
10519
10520	/*
10521	 * Run forward from the head of the blocked queue, checking each
10522	 * entry against the I/Os prior to it on the OOA queue to see if
10523	 * there is still any blockage.
10524	 *
10525	 * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10526	 * with our removing a variable on it while it is traversing the
10527	 * list.
10528	 */
10529	for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10530	     cur_blocked != NULL; cur_blocked = next_blocked) {
10531		union ctl_io *prev_ooa;
10532		ctl_action action;
10533
10534		next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10535							  blocked_links);
10536
10537		prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10538						      ctl_ooaq, ooa_links);
10539
10540		/*
10541		 * If cur_blocked happens to be the first item in the OOA
10542		 * queue now, prev_ooa will be NULL, and the action
10543		 * returned will just be CTL_ACTION_PASS.
10544		 */
10545		action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10546
10547		switch (action) {
10548		case CTL_ACTION_BLOCK:
10549			/* Nothing to do here, still blocked */
10550			break;
10551		case CTL_ACTION_OVERLAP:
10552		case CTL_ACTION_OVERLAP_TAG:
10553			/*
10554			 * This shouldn't happen!  In theory we've already
10555			 * checked this command for overlap...
10556			 */
10557			break;
10558		case CTL_ACTION_PASS:
10559		case CTL_ACTION_SKIP: {
10560			struct ctl_softc *softc;
10561			struct ctl_cmd_entry *entry;
10562			uint32_t initidx;
10563			uint8_t opcode;
10564			int isc_retval;
10565
10566			/*
10567			 * The skip case shouldn't happen, this transaction
10568			 * should have never made it onto the blocked queue.
10569			 */
10570			/*
10571			 * This I/O is no longer blocked, we can remove it
10572			 * from the blocked queue.  Since this is a TAILQ
10573			 * (doubly linked list), we can do O(1) removals
10574			 * from any place on the list.
10575			 */
10576			TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10577				     blocked_links);
10578			cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10579
10580			if (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC){
10581				/*
10582				 * Need to send IO back to original side to
10583				 * run
10584				 */
10585				union ctl_ha_msg msg_info;
10586
10587				msg_info.hdr.original_sc =
10588					cur_blocked->io_hdr.original_sc;
10589				msg_info.hdr.serializing_sc = cur_blocked;
10590				msg_info.hdr.msg_type = CTL_MSG_R2R;
10591				if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
10592				     &msg_info, sizeof(msg_info), 0)) >
10593				     CTL_HA_STATUS_SUCCESS) {
10594					printf("CTL:Check Blocked error from "
10595					       "ctl_ha_msg_send %d\n",
10596					       isc_retval);
10597				}
10598				break;
10599			}
10600			opcode = cur_blocked->scsiio.cdb[0];
10601			entry = &ctl_cmd_table[opcode];
10602			softc = control_softc;
10603
10604			initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus);
10605
10606			/*
10607			 * Check this I/O for LUN state changes that may
10608			 * have happened while this command was blocked.
10609			 * The LUN state may have been changed by a command
10610			 * ahead of us in the queue, so we need to re-check
10611			 * for any states that can be caused by SCSI
10612			 * commands.
10613			 */
10614			if (ctl_scsiio_lun_check(softc, lun, entry,
10615						 &cur_blocked->scsiio) == 0) {
10616				cur_blocked->io_hdr.flags |=
10617				                      CTL_FLAG_IS_WAS_ON_RTR;
10618				STAILQ_INSERT_TAIL(&lun->ctl_softc->rtr_queue,
10619						   &cur_blocked->io_hdr, links);
10620				/*
10621				 * In the non CTL_DONE_THREAD case, we need
10622				 * to wake up the work thread here.  When
10623				 * we're processing completed requests from
10624				 * the work thread context, we'll pop back
10625				 * around and end up pulling things off the
10626				 * RtR queue.  When we aren't processing
10627				 * things from the work thread context,
10628				 * though, we won't ever check the RtR queue.
10629				 * So we need to wake up the thread to clear
10630				 * things off the queue.  Otherwise this
10631				 * transaction will just sit on the RtR queue
10632				 * until a new I/O comes in.  (Which may or
10633				 * may not happen...)
10634				 */
10635#ifndef CTL_DONE_THREAD
10636				ctl_wakeup_thread();
10637#endif
10638			} else
10639				ctl_done_lock(cur_blocked, /*have_lock*/ 1);
10640			break;
10641		}
10642		default:
10643			/*
10644			 * This probably shouldn't happen -- we shouldn't
10645			 * get CTL_ACTION_ERROR, or anything else.
10646			 */
10647			break;
10648		}
10649	}
10650
10651	return (CTL_RETVAL_COMPLETE);
10652}
10653
10654/*
10655 * This routine (with one exception) checks LUN flags that can be set by
10656 * commands ahead of us in the OOA queue.  These flags have to be checked
10657 * when a command initially comes in, and when we pull a command off the
10658 * blocked queue and are preparing to execute it.  The reason we have to
10659 * check these flags for commands on the blocked queue is that the LUN
10660 * state may have been changed by a command ahead of us while we're on the
10661 * blocked queue.
10662 *
10663 * Ordering is somewhat important with these checks, so please pay
10664 * careful attention to the placement of any new checks.
10665 */
10666static int
10667ctl_scsiio_lun_check(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
10668		     struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10669{
10670	int retval;
10671
10672	retval = 0;
10673
10674	/*
10675	 * If this shelf is a secondary shelf controller, we have to reject
10676	 * any media access commands.
10677	 */
10678#if 0
10679	/* No longer needed for HA */
10680	if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0)
10681	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) {
10682		ctl_set_lun_standby(ctsio);
10683		retval = 1;
10684		goto bailout;
10685	}
10686#endif
10687
10688	/*
10689	 * Check for a reservation conflict.  If this command isn't allowed
10690	 * even on reserved LUNs, and if this initiator isn't the one who
10691	 * reserved us, reject the command with a reservation conflict.
10692	 */
10693	if ((lun->flags & CTL_LUN_RESERVED)
10694	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10695		if ((ctsio->io_hdr.nexus.initid.id != lun->rsv_nexus.initid.id)
10696		 || (ctsio->io_hdr.nexus.targ_port != lun->rsv_nexus.targ_port)
10697		 || (ctsio->io_hdr.nexus.targ_target.id !=
10698		     lun->rsv_nexus.targ_target.id)) {
10699			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10700			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10701			retval = 1;
10702			goto bailout;
10703		}
10704	}
10705
10706	if ( (lun->flags & CTL_LUN_PR_RESERVED)
10707	 && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV) == 0)) {
10708		uint32_t residx;
10709
10710		residx = ctl_get_resindex(&ctsio->io_hdr.nexus);
10711		/*
10712		 * if we aren't registered or it's a res holder type
10713		 * reservation and this isn't the res holder then set a
10714		 * conflict.
10715		 * NOTE: Commands which might be allowed on write exclusive
10716		 * type reservations are checked in the particular command
10717		 * for a conflict. Read and SSU are the only ones.
10718		 */
10719		if (!lun->per_res[residx].registered
10720		 || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10721			ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT;
10722			ctsio->io_hdr.status = CTL_SCSI_ERROR;
10723			retval = 1;
10724			goto bailout;
10725		}
10726
10727	}
10728
10729	if ((lun->flags & CTL_LUN_OFFLINE)
10730	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) {
10731		ctl_set_lun_not_ready(ctsio);
10732		retval = 1;
10733		goto bailout;
10734	}
10735
10736	/*
10737	 * If the LUN is stopped, see if this particular command is allowed
10738	 * for a stopped lun.  Otherwise, reject it with 0x04,0x02.
10739	 */
10740	if ((lun->flags & CTL_LUN_STOPPED)
10741	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10742		/* "Logical unit not ready, initializing cmd. required" */
10743		ctl_set_lun_stopped(ctsio);
10744		retval = 1;
10745		goto bailout;
10746	}
10747
10748	if ((lun->flags & CTL_LUN_INOPERABLE)
10749	 && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10750		/* "Medium format corrupted" */
10751		ctl_set_medium_format_corrupted(ctsio);
10752		retval = 1;
10753		goto bailout;
10754	}
10755
10756bailout:
10757	return (retval);
10758
10759}
10760
10761static void
10762ctl_failover_io(union ctl_io *io, int have_lock)
10763{
10764	ctl_set_busy(&io->scsiio);
10765	ctl_done_lock(io, have_lock);
10766}
10767
10768static void
10769ctl_failover(void)
10770{
10771	struct ctl_lun *lun;
10772	struct ctl_softc *ctl_softc;
10773	union ctl_io *next_io, *pending_io;
10774	union ctl_io *io;
10775	int lun_idx;
10776	int i;
10777
10778	ctl_softc = control_softc;
10779
10780	mtx_lock(&ctl_softc->ctl_lock);
10781	/*
10782	 * Remove any cmds from the other SC from the rtr queue.  These
10783	 * will obviously only be for LUNs for which we're the primary.
10784	 * We can't send status or get/send data for these commands.
10785	 * Since they haven't been executed yet, we can just remove them.
10786	 * We'll either abort them or delete them below, depending on
10787	 * which HA mode we're in.
10788	 */
10789	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue);
10790	     io != NULL; io = next_io) {
10791		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
10792		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10793			STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr,
10794				      ctl_io_hdr, links);
10795	}
10796
10797	for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) {
10798		lun = ctl_softc->ctl_luns[lun_idx];
10799		if (lun==NULL)
10800			continue;
10801
10802		/*
10803		 * Processor LUNs are primary on both sides.
10804		 * XXX will this always be true?
10805		 */
10806		if (lun->be_lun->lun_type == T_PROCESSOR)
10807			continue;
10808
10809		if ((lun->flags & CTL_LUN_PRIMARY_SC)
10810		 && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10811			printf("FAILOVER: primary lun %d\n", lun_idx);
10812		        /*
10813			 * Remove all commands from the other SC. First from the
10814			 * blocked queue then from the ooa queue. Once we have
10815			 * removed them. Call ctl_check_blocked to see if there
10816			 * is anything that can run.
10817			 */
10818			for (io = (union ctl_io *)TAILQ_FIRST(
10819			     &lun->blocked_queue); io != NULL; io = next_io) {
10820
10821		        	next_io = (union ctl_io *)TAILQ_NEXT(
10822				    &io->io_hdr, blocked_links);
10823
10824				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10825					TAILQ_REMOVE(&lun->blocked_queue,
10826						     &io->io_hdr,blocked_links);
10827					io->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10828					TAILQ_REMOVE(&lun->ooa_queue,
10829						     &io->io_hdr, ooa_links);
10830
10831					ctl_free_io(io);
10832				}
10833			}
10834
10835			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10836	     		     io != NULL; io = next_io) {
10837
10838		        	next_io = (union ctl_io *)TAILQ_NEXT(
10839				    &io->io_hdr, ooa_links);
10840
10841				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) {
10842
10843					TAILQ_REMOVE(&lun->ooa_queue,
10844						&io->io_hdr,
10845					     	ooa_links);
10846
10847					ctl_free_io(io);
10848				}
10849			}
10850			ctl_check_blocked(lun);
10851		} else if ((lun->flags & CTL_LUN_PRIMARY_SC)
10852			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10853
10854			printf("FAILOVER: primary lun %d\n", lun_idx);
10855			/*
10856			 * Abort all commands from the other SC.  We can't
10857			 * send status back for them now.  These should get
10858			 * cleaned up when they are completed or come out
10859			 * for a datamove operation.
10860			 */
10861			for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
10862	     		     io != NULL; io = next_io) {
10863		        	next_io = (union ctl_io *)TAILQ_NEXT(
10864					&io->io_hdr, ooa_links);
10865
10866				if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
10867					io->io_hdr.flags |= CTL_FLAG_ABORT;
10868			}
10869		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10870			&& (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) {
10871
10872			printf("FAILOVER: secondary lun %d\n", lun_idx);
10873
10874			lun->flags |= CTL_LUN_PRIMARY_SC;
10875
10876			/*
10877			 * We send all I/O that was sent to this controller
10878			 * and redirected to the other side back with
10879			 * busy status, and have the initiator retry it.
10880			 * Figuring out how much data has been transferred,
10881			 * etc. and picking up where we left off would be
10882			 * very tricky.
10883			 *
10884			 * XXX KDM need to remove I/O from the blocked
10885			 * queue as well!
10886			 */
10887			for (pending_io = (union ctl_io *)TAILQ_FIRST(
10888			     &lun->ooa_queue); pending_io != NULL;
10889			     pending_io = next_io) {
10890
10891				next_io =  (union ctl_io *)TAILQ_NEXT(
10892					&pending_io->io_hdr, ooa_links);
10893
10894				pending_io->io_hdr.flags &=
10895					~CTL_FLAG_SENT_2OTHER_SC;
10896
10897				if (pending_io->io_hdr.flags &
10898				    CTL_FLAG_IO_ACTIVE) {
10899					pending_io->io_hdr.flags |=
10900						CTL_FLAG_FAILOVER;
10901				} else {
10902					ctl_set_busy(&pending_io->scsiio);
10903					ctl_done_lock(pending_io,
10904						      /*have_lock*/1);
10905				}
10906			}
10907
10908			/*
10909			 * Build Unit Attention
10910			 */
10911			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
10912				lun->pending_sense[i].ua_pending |=
10913				                     CTL_UA_ASYM_ACC_CHANGE;
10914			}
10915		} else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0)
10916			&& (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) {
10917			printf("FAILOVER: secondary lun %d\n", lun_idx);
10918			/*
10919			 * if the first io on the OOA is not on the RtR queue
10920			 * add it.
10921			 */
10922			lun->flags |= CTL_LUN_PRIMARY_SC;
10923
10924			pending_io = (union ctl_io *)TAILQ_FIRST(
10925			    &lun->ooa_queue);
10926			if (pending_io==NULL) {
10927				printf("Nothing on OOA queue\n");
10928				continue;
10929			}
10930
10931			pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10932			if ((pending_io->io_hdr.flags &
10933			     CTL_FLAG_IS_WAS_ON_RTR) == 0) {
10934				pending_io->io_hdr.flags |=
10935				    CTL_FLAG_IS_WAS_ON_RTR;
10936				STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
10937						   &pending_io->io_hdr, links);
10938			}
10939#if 0
10940			else
10941			{
10942				printf("Tag 0x%04x is running\n",
10943				      pending_io->scsiio.tag_num);
10944			}
10945#endif
10946
10947			next_io = (union ctl_io *)TAILQ_NEXT(
10948			    &pending_io->io_hdr, ooa_links);
10949			for (pending_io=next_io; pending_io != NULL;
10950			     pending_io = next_io) {
10951				pending_io->io_hdr.flags &=
10952				    ~CTL_FLAG_SENT_2OTHER_SC;
10953				next_io = (union ctl_io *)TAILQ_NEXT(
10954					&pending_io->io_hdr, ooa_links);
10955				if (pending_io->io_hdr.flags &
10956				    CTL_FLAG_IS_WAS_ON_RTR) {
10957#if 0
10958				        printf("Tag 0x%04x is running\n",
10959				      		pending_io->scsiio.tag_num);
10960#endif
10961					continue;
10962				}
10963
10964				switch (ctl_check_ooa(lun, pending_io,
10965			            (union ctl_io *)TAILQ_PREV(
10966				    &pending_io->io_hdr, ctl_ooaq,
10967				    ooa_links))) {
10968
10969				case CTL_ACTION_BLOCK:
10970					TAILQ_INSERT_TAIL(&lun->blocked_queue,
10971							  &pending_io->io_hdr,
10972							  blocked_links);
10973					pending_io->io_hdr.flags |=
10974					    CTL_FLAG_BLOCKED;
10975					break;
10976				case CTL_ACTION_PASS:
10977				case CTL_ACTION_SKIP:
10978					pending_io->io_hdr.flags |=
10979					    CTL_FLAG_IS_WAS_ON_RTR;
10980					STAILQ_INSERT_TAIL(
10981					    &ctl_softc->rtr_queue,
10982					    &pending_io->io_hdr, links);
10983					break;
10984				case CTL_ACTION_OVERLAP:
10985					ctl_set_overlapped_cmd(
10986					    (struct ctl_scsiio *)pending_io);
10987					ctl_done_lock(pending_io,
10988						      /*have_lock*/ 1);
10989					break;
10990				case CTL_ACTION_OVERLAP_TAG:
10991					ctl_set_overlapped_tag(
10992					    (struct ctl_scsiio *)pending_io,
10993					    pending_io->scsiio.tag_num & 0xff);
10994					ctl_done_lock(pending_io,
10995						      /*have_lock*/ 1);
10996					break;
10997				case CTL_ACTION_ERROR:
10998				default:
10999					ctl_set_internal_failure(
11000						(struct ctl_scsiio *)pending_io,
11001						0,  // sks_valid
11002						0); //retry count
11003					ctl_done_lock(pending_io,
11004						      /*have_lock*/ 1);
11005					break;
11006				}
11007			}
11008
11009			/*
11010			 * Build Unit Attention
11011			 */
11012			for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11013				lun->pending_sense[i].ua_pending |=
11014				                     CTL_UA_ASYM_ACC_CHANGE;
11015			}
11016		} else {
11017			panic("Unhandled HA mode failover, LUN flags = %#x, "
11018			      "ha_mode = #%x", lun->flags, ctl_softc->ha_mode);
11019		}
11020	}
11021	ctl_pause_rtr = 0;
11022	mtx_unlock(&ctl_softc->ctl_lock);
11023}
11024
11025static int
11026ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio)
11027{
11028	struct ctl_lun *lun;
11029	struct ctl_cmd_entry *entry;
11030	uint8_t opcode;
11031	uint32_t initidx, targ_lun;
11032	int retval;
11033
11034	retval = 0;
11035
11036	lun = NULL;
11037
11038	opcode = ctsio->cdb[0];
11039
11040	mtx_lock(&ctl_softc->ctl_lock);
11041
11042	targ_lun = ctsio->io_hdr.nexus.targ_lun;
11043	if (ctsio->io_hdr.nexus.lun_map_fn != NULL)
11044		targ_lun = ctsio->io_hdr.nexus.lun_map_fn(ctsio->io_hdr.nexus.lun_map_arg, targ_lun);
11045	if ((targ_lun < CTL_MAX_LUNS)
11046	 && (ctl_softc->ctl_luns[targ_lun] != NULL)) {
11047		lun = ctl_softc->ctl_luns[targ_lun];
11048		/*
11049		 * If the LUN is invalid, pretend that it doesn't exist.
11050		 * It will go away as soon as all pending I/O has been
11051		 * completed.
11052		 */
11053		if (lun->flags & CTL_LUN_DISABLED) {
11054			lun = NULL;
11055		} else {
11056			ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11057			ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11058				lun->be_lun;
11059			if (lun->be_lun->lun_type == T_PROCESSOR) {
11060				ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV;
11061			}
11062		}
11063	} else {
11064		ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11065		ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11066	}
11067
11068	entry = &ctl_cmd_table[opcode];
11069
11070	ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11071	ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11072
11073	/*
11074	 * Check to see whether we can send this command to LUNs that don't
11075	 * exist.  This should pretty much only be the case for inquiry
11076	 * and request sense.  Further checks, below, really require having
11077	 * a LUN, so we can't really check the command anymore.  Just put
11078	 * it on the rtr queue.
11079	 */
11080	if (lun == NULL) {
11081		if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11082			goto queue_rtr;
11083
11084		ctl_set_unsupported_lun(ctsio);
11085		mtx_unlock(&ctl_softc->ctl_lock);
11086		ctl_done((union ctl_io *)ctsio);
11087		CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11088		goto bailout;
11089	} else {
11090		/*
11091		 * Every I/O goes into the OOA queue for a particular LUN, and
11092		 * stays there until completion.
11093		 */
11094		TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11095
11096		/*
11097		 * Make sure we support this particular command on this LUN.
11098		 * e.g., we don't support writes to the control LUN.
11099		 */
11100		switch (lun->be_lun->lun_type) {
11101		case T_PROCESSOR:
11102		 	if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11103			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11104			      == 0)) {
11105				ctl_set_invalid_opcode(ctsio);
11106				mtx_unlock(&ctl_softc->ctl_lock);
11107				ctl_done((union ctl_io *)ctsio);
11108				goto bailout;
11109			}
11110			break;
11111		case T_DIRECT:
11112			if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11113			 && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS)
11114			      == 0)){
11115				ctl_set_invalid_opcode(ctsio);
11116				mtx_unlock(&ctl_softc->ctl_lock);
11117				ctl_done((union ctl_io *)ctsio);
11118				goto bailout;
11119			}
11120			break;
11121		default:
11122			printf("Unsupported CTL LUN type %d\n",
11123			       lun->be_lun->lun_type);
11124			panic("Unsupported CTL LUN type %d\n",
11125			      lun->be_lun->lun_type);
11126			break; /* NOTREACHED */
11127		}
11128	}
11129
11130	initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11131
11132	/*
11133	 * If we've got a request sense, it'll clear the contingent
11134	 * allegiance condition.  Otherwise, if we have a CA condition for
11135	 * this initiator, clear it, because it sent down a command other
11136	 * than request sense.
11137	 */
11138	if ((opcode != REQUEST_SENSE)
11139	 && (ctl_is_set(lun->have_ca, initidx)))
11140		ctl_clear_mask(lun->have_ca, initidx);
11141
11142	/*
11143	 * If the command has this flag set, it handles its own unit
11144	 * attention reporting, we shouldn't do anything.  Otherwise we
11145	 * check for any pending unit attentions, and send them back to the
11146	 * initiator.  We only do this when a command initially comes in,
11147	 * not when we pull it off the blocked queue.
11148	 *
11149	 * According to SAM-3, section 5.3.2, the order that things get
11150	 * presented back to the host is basically unit attentions caused
11151	 * by some sort of reset event, busy status, reservation conflicts
11152	 * or task set full, and finally any other status.
11153	 *
11154	 * One issue here is that some of the unit attentions we report
11155	 * don't fall into the "reset" category (e.g. "reported luns data
11156	 * has changed").  So reporting it here, before the reservation
11157	 * check, may be technically wrong.  I guess the only thing to do
11158	 * would be to check for and report the reset events here, and then
11159	 * check for the other unit attention types after we check for a
11160	 * reservation conflict.
11161	 *
11162	 * XXX KDM need to fix this
11163	 */
11164	if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11165		ctl_ua_type ua_type;
11166
11167		ua_type = lun->pending_sense[initidx].ua_pending;
11168		if (ua_type != CTL_UA_NONE) {
11169			scsi_sense_data_type sense_format;
11170
11171			if (lun != NULL)
11172				sense_format = (lun->flags &
11173				    CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC :
11174				    SSD_TYPE_FIXED;
11175			else
11176				sense_format = SSD_TYPE_FIXED;
11177
11178			ua_type = ctl_build_ua(ua_type, &ctsio->sense_data,
11179					       sense_format);
11180			if (ua_type != CTL_UA_NONE) {
11181				ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11182				ctsio->io_hdr.status = CTL_SCSI_ERROR |
11183						       CTL_AUTOSENSE;
11184				ctsio->sense_len = SSD_FULL_SIZE;
11185				lun->pending_sense[initidx].ua_pending &=
11186					~ua_type;
11187				mtx_unlock(&ctl_softc->ctl_lock);
11188				ctl_done((union ctl_io *)ctsio);
11189				goto bailout;
11190			}
11191		}
11192	}
11193
11194
11195	if (ctl_scsiio_lun_check(ctl_softc, lun, entry, ctsio) != 0) {
11196		mtx_unlock(&ctl_softc->ctl_lock);
11197		ctl_done((union ctl_io *)ctsio);
11198		goto bailout;
11199	}
11200
11201	/*
11202	 * XXX CHD this is where we want to send IO to other side if
11203	 * this LUN is secondary on this SC. We will need to make a copy
11204	 * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11205	 * the copy we send as FROM_OTHER.
11206	 * We also need to stuff the address of the original IO so we can
11207	 * find it easily. Something similar will need be done on the other
11208	 * side so when we are done we can find the copy.
11209	 */
11210	if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11211		union ctl_ha_msg msg_info;
11212		int isc_retval;
11213
11214		ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11215
11216		msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11217		msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11218#if 0
11219		printf("1. ctsio %p\n", ctsio);
11220#endif
11221		msg_info.hdr.serializing_sc = NULL;
11222		msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11223		msg_info.scsi.tag_num = ctsio->tag_num;
11224		msg_info.scsi.tag_type = ctsio->tag_type;
11225		memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11226
11227		ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11228
11229		if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11230		    (void *)&msg_info, sizeof(msg_info), 0)) >
11231		    CTL_HA_STATUS_SUCCESS) {
11232			printf("CTL:precheck, ctl_ha_msg_send returned %d\n",
11233			       isc_retval);
11234			printf("CTL:opcode is %x\n",opcode);
11235		} else {
11236#if 0
11237			printf("CTL:Precheck sent msg, opcode is %x\n",opcode);
11238#endif
11239		}
11240
11241		/*
11242		 * XXX KDM this I/O is off the incoming queue, but hasn't
11243		 * been inserted on any other queue.  We may need to come
11244		 * up with a holding queue while we wait for serialization
11245		 * so that we have an idea of what we're waiting for from
11246		 * the other side.
11247		 */
11248		goto bailout_unlock;
11249	}
11250
11251	switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11252			      (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11253			      ctl_ooaq, ooa_links))) {
11254	case CTL_ACTION_BLOCK:
11255		ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11256		TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11257				  blocked_links);
11258		goto bailout_unlock;
11259		break; /* NOTREACHED */
11260	case CTL_ACTION_PASS:
11261	case CTL_ACTION_SKIP:
11262		goto queue_rtr;
11263		break; /* NOTREACHED */
11264	case CTL_ACTION_OVERLAP:
11265		ctl_set_overlapped_cmd(ctsio);
11266		mtx_unlock(&ctl_softc->ctl_lock);
11267		ctl_done((union ctl_io *)ctsio);
11268		goto bailout;
11269		break; /* NOTREACHED */
11270	case CTL_ACTION_OVERLAP_TAG:
11271		ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11272		mtx_unlock(&ctl_softc->ctl_lock);
11273		ctl_done((union ctl_io *)ctsio);
11274		goto bailout;
11275		break; /* NOTREACHED */
11276	case CTL_ACTION_ERROR:
11277	default:
11278		ctl_set_internal_failure(ctsio,
11279					 /*sks_valid*/ 0,
11280					 /*retry_count*/ 0);
11281		mtx_unlock(&ctl_softc->ctl_lock);
11282		ctl_done((union ctl_io *)ctsio);
11283		goto bailout;
11284		break; /* NOTREACHED */
11285	}
11286
11287	goto bailout_unlock;
11288
11289queue_rtr:
11290	ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11291	STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue, &ctsio->io_hdr, links);
11292
11293bailout_unlock:
11294	mtx_unlock(&ctl_softc->ctl_lock);
11295
11296bailout:
11297	return (retval);
11298}
11299
11300static int
11301ctl_scsiio(struct ctl_scsiio *ctsio)
11302{
11303	int retval;
11304	struct ctl_cmd_entry *entry;
11305
11306	retval = CTL_RETVAL_COMPLETE;
11307
11308	CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11309
11310	entry = &ctl_cmd_table[ctsio->cdb[0]];
11311
11312	/*
11313	 * If this I/O has been aborted, just send it straight to
11314	 * ctl_done() without executing it.
11315	 */
11316	if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11317		ctl_done((union ctl_io *)ctsio);
11318		goto bailout;
11319	}
11320
11321	/*
11322	 * All the checks should have been handled by ctl_scsiio_precheck().
11323	 * We should be clear now to just execute the I/O.
11324	 */
11325	retval = entry->execute(ctsio);
11326
11327bailout:
11328	return (retval);
11329}
11330
11331/*
11332 * Since we only implement one target right now, a bus reset simply resets
11333 * our single target.
11334 */
11335static int
11336ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io)
11337{
11338	return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET));
11339}
11340
11341static int
11342ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
11343		 ctl_ua_type ua_type)
11344{
11345	struct ctl_lun *lun;
11346	int retval;
11347
11348	if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11349		union ctl_ha_msg msg_info;
11350
11351		io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11352		msg_info.hdr.nexus = io->io_hdr.nexus;
11353		if (ua_type==CTL_UA_TARG_RESET)
11354			msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11355		else
11356			msg_info.task.task_action = CTL_TASK_BUS_RESET;
11357		msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11358		msg_info.hdr.original_sc = NULL;
11359		msg_info.hdr.serializing_sc = NULL;
11360		if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11361		    (void *)&msg_info, sizeof(msg_info), 0)) {
11362		}
11363	}
11364	retval = 0;
11365
11366	STAILQ_FOREACH(lun, &ctl_softc->lun_list, links)
11367		retval += ctl_lun_reset(lun, io, ua_type);
11368
11369	return (retval);
11370}
11371
11372/*
11373 * The LUN should always be set.  The I/O is optional, and is used to
11374 * distinguish between I/Os sent by this initiator, and by other
11375 * initiators.  We set unit attention for initiators other than this one.
11376 * SAM-3 is vague on this point.  It does say that a unit attention should
11377 * be established for other initiators when a LUN is reset (see section
11378 * 5.7.3), but it doesn't specifically say that the unit attention should
11379 * be established for this particular initiator when a LUN is reset.  Here
11380 * is the relevant text, from SAM-3 rev 8:
11381 *
11382 * 5.7.2 When a SCSI initiator port aborts its own tasks
11383 *
11384 * When a SCSI initiator port causes its own task(s) to be aborted, no
11385 * notification that the task(s) have been aborted shall be returned to
11386 * the SCSI initiator port other than the completion response for the
11387 * command or task management function action that caused the task(s) to
11388 * be aborted and notification(s) associated with related effects of the
11389 * action (e.g., a reset unit attention condition).
11390 *
11391 * XXX KDM for now, we're setting unit attention for all initiators.
11392 */
11393static int
11394ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11395{
11396	union ctl_io *xio;
11397#if 0
11398	uint32_t initindex;
11399#endif
11400	int i;
11401
11402	/*
11403	 * Run through the OOA queue and abort each I/O.
11404	 */
11405#if 0
11406	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11407#endif
11408	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11409	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11410		xio->io_hdr.flags |= CTL_FLAG_ABORT;
11411	}
11412
11413	/*
11414	 * This version sets unit attention for every
11415	 */
11416#if 0
11417	initindex = ctl_get_initindex(&io->io_hdr.nexus);
11418	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11419		if (initindex == i)
11420			continue;
11421		lun->pending_sense[i].ua_pending |= ua_type;
11422	}
11423#endif
11424
11425	/*
11426	 * A reset (any kind, really) clears reservations established with
11427	 * RESERVE/RELEASE.  It does not clear reservations established
11428	 * with PERSISTENT RESERVE OUT, but we don't support that at the
11429	 * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11430	 * reservations made with the RESERVE/RELEASE commands, because
11431	 * those commands are obsolete in SPC-3.
11432	 */
11433	lun->flags &= ~CTL_LUN_RESERVED;
11434
11435	for (i = 0; i < CTL_MAX_INITIATORS; i++) {
11436		ctl_clear_mask(lun->have_ca, i);
11437		lun->pending_sense[i].ua_pending |= ua_type;
11438	}
11439
11440	return (0);
11441}
11442
11443static int
11444ctl_abort_task(union ctl_io *io)
11445{
11446	union ctl_io *xio;
11447	struct ctl_lun *lun;
11448	struct ctl_softc *ctl_softc;
11449#if 0
11450	struct sbuf sb;
11451	char printbuf[128];
11452#endif
11453	int found;
11454	uint32_t targ_lun;
11455
11456	ctl_softc = control_softc;
11457	found = 0;
11458
11459	/*
11460	 * Look up the LUN.
11461	 */
11462	targ_lun = io->io_hdr.nexus.targ_lun;
11463	if (io->io_hdr.nexus.lun_map_fn != NULL)
11464		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11465	if ((targ_lun < CTL_MAX_LUNS)
11466	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11467		lun = ctl_softc->ctl_luns[targ_lun];
11468	else
11469		goto bailout;
11470
11471#if 0
11472	printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11473	       lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11474#endif
11475
11476	/*
11477	 * Run through the OOA queue and attempt to find the given I/O.
11478	 * The target port, initiator ID, tag type and tag number have to
11479	 * match the values that we got from the initiator.  If we have an
11480	 * untagged command to abort, simply abort the first untagged command
11481	 * we come to.  We only allow one untagged command at a time of course.
11482	 */
11483#if 0
11484	TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) {
11485#endif
11486	for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11487	     xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11488#if 0
11489		sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11490
11491		sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11492			    lun->lun, xio->scsiio.tag_num,
11493			    xio->scsiio.tag_type,
11494			    (xio->io_hdr.blocked_links.tqe_prev
11495			    == NULL) ? "" : " BLOCKED",
11496			    (xio->io_hdr.flags &
11497			    CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11498			    (xio->io_hdr.flags &
11499			    CTL_FLAG_ABORT) ? " ABORT" : "",
11500			    (xio->io_hdr.flags &
11501			    CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11502		ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11503		sbuf_finish(&sb);
11504		printf("%s\n", sbuf_data(&sb));
11505#endif
11506
11507		if ((xio->io_hdr.nexus.targ_port == io->io_hdr.nexus.targ_port)
11508		 && (xio->io_hdr.nexus.initid.id ==
11509		     io->io_hdr.nexus.initid.id)) {
11510			/*
11511			 * If the abort says that the task is untagged, the
11512			 * task in the queue must be untagged.  Otherwise,
11513			 * we just check to see whether the tag numbers
11514			 * match.  This is because the QLogic firmware
11515			 * doesn't pass back the tag type in an abort
11516			 * request.
11517			 */
11518#if 0
11519			if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11520			  && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11521			 || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11522#endif
11523			/*
11524			 * XXX KDM we've got problems with FC, because it
11525			 * doesn't send down a tag type with aborts.  So we
11526			 * can only really go by the tag number...
11527			 * This may cause problems with parallel SCSI.
11528			 * Need to figure that out!!
11529			 */
11530			if (xio->scsiio.tag_num == io->taskio.tag_num) {
11531				xio->io_hdr.flags |= CTL_FLAG_ABORT;
11532				found = 1;
11533				if ((io->io_hdr.flags &
11534				     CTL_FLAG_FROM_OTHER_SC) == 0 &&
11535				    !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11536					union ctl_ha_msg msg_info;
11537
11538					io->io_hdr.flags |=
11539					                CTL_FLAG_SENT_2OTHER_SC;
11540					msg_info.hdr.nexus = io->io_hdr.nexus;
11541					msg_info.task.task_action =
11542						CTL_TASK_ABORT_TASK;
11543					msg_info.task.tag_num =
11544						io->taskio.tag_num;
11545					msg_info.task.tag_type =
11546						io->taskio.tag_type;
11547					msg_info.hdr.msg_type =
11548						CTL_MSG_MANAGE_TASKS;
11549					msg_info.hdr.original_sc = NULL;
11550					msg_info.hdr.serializing_sc = NULL;
11551#if 0
11552					printf("Sent Abort to other side\n");
11553#endif
11554					if (CTL_HA_STATUS_SUCCESS !=
11555					        ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11556		    				(void *)&msg_info,
11557						sizeof(msg_info), 0)) {
11558					}
11559				}
11560#if 0
11561				printf("ctl_abort_task: found I/O to abort\n");
11562#endif
11563				break;
11564			}
11565		}
11566	}
11567
11568bailout:
11569
11570	if (found == 0) {
11571		/*
11572		 * This isn't really an error.  It's entirely possible for
11573		 * the abort and command completion to cross on the wire.
11574		 * This is more of an informative/diagnostic error.
11575		 */
11576#if 0
11577		printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11578		       "%d:%d:%d:%d tag %d type %d\n",
11579		       io->io_hdr.nexus.initid.id,
11580		       io->io_hdr.nexus.targ_port,
11581		       io->io_hdr.nexus.targ_target.id,
11582		       io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11583		       io->taskio.tag_type);
11584#endif
11585		return (1);
11586	} else
11587		return (0);
11588}
11589
11590/*
11591 * This routine cannot block!  It must be callable from an interrupt
11592 * handler as well as from the work thread.
11593 */
11594static void
11595ctl_run_task_queue(struct ctl_softc *ctl_softc)
11596{
11597	union ctl_io *io, *next_io;
11598
11599	mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
11600
11601	CTL_DEBUG_PRINT(("ctl_run_task_queue\n"));
11602
11603	for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->task_queue);
11604	     io != NULL; io = next_io) {
11605		int retval;
11606		const char *task_desc;
11607
11608		next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links);
11609
11610		retval = 0;
11611
11612		switch (io->io_hdr.io_type) {
11613		case CTL_IO_TASK: {
11614			task_desc = ctl_scsi_task_string(&io->taskio);
11615			if (task_desc != NULL) {
11616#ifdef NEEDTOPORT
11617				csevent_log(CSC_CTL | CSC_SHELF_SW |
11618					    CTL_TASK_REPORT,
11619					    csevent_LogType_Trace,
11620					    csevent_Severity_Information,
11621					    csevent_AlertLevel_Green,
11622					    csevent_FRU_Firmware,
11623					    csevent_FRU_Unknown,
11624					    "CTL: received task: %s",task_desc);
11625#endif
11626			} else {
11627#ifdef NEEDTOPORT
11628				csevent_log(CSC_CTL | CSC_SHELF_SW |
11629					    CTL_TASK_REPORT,
11630					    csevent_LogType_Trace,
11631					    csevent_Severity_Information,
11632					    csevent_AlertLevel_Green,
11633					    csevent_FRU_Firmware,
11634					    csevent_FRU_Unknown,
11635					    "CTL: received unknown task "
11636					    "type: %d (%#x)",
11637					    io->taskio.task_action,
11638					    io->taskio.task_action);
11639#endif
11640			}
11641			switch (io->taskio.task_action) {
11642			case CTL_TASK_ABORT_TASK:
11643				retval = ctl_abort_task(io);
11644				break;
11645			case CTL_TASK_ABORT_TASK_SET:
11646				break;
11647			case CTL_TASK_CLEAR_ACA:
11648				break;
11649			case CTL_TASK_CLEAR_TASK_SET:
11650				break;
11651			case CTL_TASK_LUN_RESET: {
11652				struct ctl_lun *lun;
11653				uint32_t targ_lun;
11654				int retval;
11655
11656				targ_lun = io->io_hdr.nexus.targ_lun;
11657				if (io->io_hdr.nexus.lun_map_fn != NULL)
11658					targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11659
11660				if ((targ_lun < CTL_MAX_LUNS)
11661				 && (ctl_softc->ctl_luns[targ_lun] != NULL))
11662					lun = ctl_softc->ctl_luns[targ_lun];
11663				else {
11664					retval = 1;
11665					break;
11666				}
11667
11668				if (!(io->io_hdr.flags &
11669				    CTL_FLAG_FROM_OTHER_SC)) {
11670					union ctl_ha_msg msg_info;
11671
11672					io->io_hdr.flags |=
11673						CTL_FLAG_SENT_2OTHER_SC;
11674					msg_info.hdr.msg_type =
11675						CTL_MSG_MANAGE_TASKS;
11676					msg_info.hdr.nexus = io->io_hdr.nexus;
11677					msg_info.task.task_action =
11678						CTL_TASK_LUN_RESET;
11679					msg_info.hdr.original_sc = NULL;
11680					msg_info.hdr.serializing_sc = NULL;
11681					if (CTL_HA_STATUS_SUCCESS !=
11682					    ctl_ha_msg_send(CTL_HA_CHAN_CTL,
11683					    (void *)&msg_info,
11684					    sizeof(msg_info), 0)) {
11685					}
11686				}
11687
11688				retval = ctl_lun_reset(lun, io,
11689						       CTL_UA_LUN_RESET);
11690				break;
11691			}
11692			case CTL_TASK_TARGET_RESET:
11693				retval = ctl_target_reset(ctl_softc, io,
11694							  CTL_UA_TARG_RESET);
11695				break;
11696			case CTL_TASK_BUS_RESET:
11697				retval = ctl_bus_reset(ctl_softc, io);
11698				break;
11699			case CTL_TASK_PORT_LOGIN:
11700				break;
11701			case CTL_TASK_PORT_LOGOUT:
11702				break;
11703			default:
11704				printf("ctl_run_task_queue: got unknown task "
11705				       "management event %d\n",
11706				       io->taskio.task_action);
11707				break;
11708			}
11709			if (retval == 0)
11710				io->io_hdr.status = CTL_SUCCESS;
11711			else
11712				io->io_hdr.status = CTL_ERROR;
11713
11714			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11715				      ctl_io_hdr, links);
11716			/*
11717			 * This will queue this I/O to the done queue, but the
11718			 * work thread won't be able to process it until we
11719			 * return and the lock is released.
11720			 */
11721			ctl_done_lock(io, /*have_lock*/ 1);
11722			break;
11723		}
11724		default: {
11725
11726			printf("%s: invalid I/O type %d msg %d cdb %x"
11727			       " iptl: %ju:%d:%ju:%d tag 0x%04x\n",
11728			       __func__, io->io_hdr.io_type,
11729			       io->io_hdr.msg_type, io->scsiio.cdb[0],
11730			       (uintmax_t)io->io_hdr.nexus.initid.id,
11731			       io->io_hdr.nexus.targ_port,
11732			       (uintmax_t)io->io_hdr.nexus.targ_target.id,
11733			       io->io_hdr.nexus.targ_lun /* XXX */,
11734			       (io->io_hdr.io_type == CTL_IO_TASK) ?
11735			       io->taskio.tag_num : io->scsiio.tag_num);
11736			STAILQ_REMOVE(&ctl_softc->task_queue, &io->io_hdr,
11737				      ctl_io_hdr, links);
11738			ctl_free_io(io);
11739			break;
11740		}
11741		}
11742	}
11743
11744	ctl_softc->flags &= ~CTL_FLAG_TASK_PENDING;
11745}
11746
11747/*
11748 * For HA operation.  Handle commands that come in from the other
11749 * controller.
11750 */
11751static void
11752ctl_handle_isc(union ctl_io *io)
11753{
11754	int free_io;
11755	struct ctl_lun *lun;
11756	struct ctl_softc *ctl_softc;
11757	uint32_t targ_lun;
11758
11759	ctl_softc = control_softc;
11760
11761	targ_lun = io->io_hdr.nexus.targ_lun;
11762	if (io->io_hdr.nexus.lun_map_fn != NULL)
11763		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
11764	lun = ctl_softc->ctl_luns[targ_lun];
11765
11766	switch (io->io_hdr.msg_type) {
11767	case CTL_MSG_SERIALIZE:
11768		free_io = ctl_serialize_other_sc_cmd(&io->scsiio,
11769						     /*have_lock*/ 0);
11770		break;
11771	case CTL_MSG_R2R: {
11772		uint8_t opcode;
11773		struct ctl_cmd_entry *entry;
11774
11775		/*
11776		 * This is only used in SER_ONLY mode.
11777		 */
11778		free_io = 0;
11779		opcode = io->scsiio.cdb[0];
11780		entry = &ctl_cmd_table[opcode];
11781		mtx_lock(&ctl_softc->ctl_lock);
11782		if (ctl_scsiio_lun_check(ctl_softc, lun,
11783		    entry, (struct ctl_scsiio *)io) != 0) {
11784			ctl_done_lock(io, /*have_lock*/ 1);
11785			mtx_unlock(&ctl_softc->ctl_lock);
11786			break;
11787		}
11788		io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11789		STAILQ_INSERT_TAIL(&ctl_softc->rtr_queue,
11790				   &io->io_hdr, links);
11791		mtx_unlock(&ctl_softc->ctl_lock);
11792		break;
11793	}
11794	case CTL_MSG_FINISH_IO:
11795		if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
11796			free_io = 0;
11797			ctl_done_lock(io, /*have_lock*/ 0);
11798		} else {
11799			free_io = 1;
11800			mtx_lock(&ctl_softc->ctl_lock);
11801			TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11802				     ooa_links);
11803			STAILQ_REMOVE(&ctl_softc->task_queue,
11804				      &io->io_hdr, ctl_io_hdr, links);
11805			ctl_check_blocked(lun);
11806			mtx_unlock(&ctl_softc->ctl_lock);
11807		}
11808		break;
11809	case CTL_MSG_PERS_ACTION:
11810		ctl_hndl_per_res_out_on_other_sc(
11811			(union ctl_ha_msg *)&io->presio.pr_msg);
11812		free_io = 1;
11813		break;
11814	case CTL_MSG_BAD_JUJU:
11815		free_io = 0;
11816		ctl_done_lock(io, /*have_lock*/ 0);
11817		break;
11818	case CTL_MSG_DATAMOVE:
11819		/* Only used in XFER mode */
11820		free_io = 0;
11821		ctl_datamove_remote(io);
11822		break;
11823	case CTL_MSG_DATAMOVE_DONE:
11824		/* Only used in XFER mode */
11825		free_io = 0;
11826		io->scsiio.be_move_done(io);
11827		break;
11828	default:
11829		free_io = 1;
11830		printf("%s: Invalid message type %d\n",
11831		       __func__, io->io_hdr.msg_type);
11832		break;
11833	}
11834	if (free_io)
11835		ctl_free_io(io);
11836
11837}
11838
11839
11840/*
11841 * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11842 * there is no match.
11843 */
11844static ctl_lun_error_pattern
11845ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11846{
11847	struct ctl_cmd_entry *entry;
11848	ctl_lun_error_pattern filtered_pattern, pattern;
11849	uint8_t opcode;
11850
11851	pattern = desc->error_pattern;
11852
11853	/*
11854	 * XXX KDM we need more data passed into this function to match a
11855	 * custom pattern, and we actually need to implement custom pattern
11856	 * matching.
11857	 */
11858	if (pattern & CTL_LUN_PAT_CMD)
11859		return (CTL_LUN_PAT_CMD);
11860
11861	if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11862		return (CTL_LUN_PAT_ANY);
11863
11864	opcode = ctsio->cdb[0];
11865	entry = &ctl_cmd_table[opcode];
11866
11867	filtered_pattern = entry->pattern & pattern;
11868
11869	/*
11870	 * If the user requested specific flags in the pattern (e.g.
11871	 * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11872	 * flags.
11873	 *
11874	 * If the user did not specify any flags, it doesn't matter whether
11875	 * or not the command supports the flags.
11876	 */
11877	if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11878	     (pattern & ~CTL_LUN_PAT_MASK))
11879		return (CTL_LUN_PAT_NONE);
11880
11881	/*
11882	 * If the user asked for a range check, see if the requested LBA
11883	 * range overlaps with this command's LBA range.
11884	 */
11885	if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11886		uint64_t lba1;
11887		uint32_t len1;
11888		ctl_action action;
11889		int retval;
11890
11891		retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11892		if (retval != 0)
11893			return (CTL_LUN_PAT_NONE);
11894
11895		action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11896					      desc->lba_range.len);
11897		/*
11898		 * A "pass" means that the LBA ranges don't overlap, so
11899		 * this doesn't match the user's range criteria.
11900		 */
11901		if (action == CTL_ACTION_PASS)
11902			return (CTL_LUN_PAT_NONE);
11903	}
11904
11905	return (filtered_pattern);
11906}
11907
11908static void
11909ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11910{
11911	struct ctl_error_desc *desc, *desc2;
11912
11913	mtx_assert(&control_softc->ctl_lock, MA_OWNED);
11914
11915	STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11916		ctl_lun_error_pattern pattern;
11917		/*
11918		 * Check to see whether this particular command matches
11919		 * the pattern in the descriptor.
11920		 */
11921		pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11922		if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11923			continue;
11924
11925		switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11926		case CTL_LUN_INJ_ABORTED:
11927			ctl_set_aborted(&io->scsiio);
11928			break;
11929		case CTL_LUN_INJ_MEDIUM_ERR:
11930			ctl_set_medium_error(&io->scsiio);
11931			break;
11932		case CTL_LUN_INJ_UA:
11933			/* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11934			 * OCCURRED */
11935			ctl_set_ua(&io->scsiio, 0x29, 0x00);
11936			break;
11937		case CTL_LUN_INJ_CUSTOM:
11938			/*
11939			 * We're assuming the user knows what he is doing.
11940			 * Just copy the sense information without doing
11941			 * checks.
11942			 */
11943			bcopy(&desc->custom_sense, &io->scsiio.sense_data,
11944			      ctl_min(sizeof(desc->custom_sense),
11945				      sizeof(io->scsiio.sense_data)));
11946			io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
11947			io->scsiio.sense_len = SSD_FULL_SIZE;
11948			io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11949			break;
11950		case CTL_LUN_INJ_NONE:
11951		default:
11952			/*
11953			 * If this is an error injection type we don't know
11954			 * about, clear the continuous flag (if it is set)
11955			 * so it will get deleted below.
11956			 */
11957			desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
11958			break;
11959		}
11960		/*
11961		 * By default, each error injection action is a one-shot
11962		 */
11963		if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
11964			continue;
11965
11966		STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
11967
11968		free(desc, M_CTL);
11969	}
11970}
11971
11972#ifdef CTL_IO_DELAY
11973static void
11974ctl_datamove_timer_wakeup(void *arg)
11975{
11976	union ctl_io *io;
11977
11978	io = (union ctl_io *)arg;
11979
11980	ctl_datamove(io);
11981}
11982#endif /* CTL_IO_DELAY */
11983
11984void
11985ctl_datamove(union ctl_io *io)
11986{
11987	void (*fe_datamove)(union ctl_io *io);
11988
11989	mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
11990
11991	CTL_DEBUG_PRINT(("ctl_datamove\n"));
11992
11993#ifdef CTL_TIME_IO
11994	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
11995		char str[256];
11996		char path_str[64];
11997		struct sbuf sb;
11998
11999		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12000		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12001
12002		sbuf_cat(&sb, path_str);
12003		switch (io->io_hdr.io_type) {
12004		case CTL_IO_SCSI:
12005			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12006			sbuf_printf(&sb, "\n");
12007			sbuf_cat(&sb, path_str);
12008			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12009				    io->scsiio.tag_num, io->scsiio.tag_type);
12010			break;
12011		case CTL_IO_TASK:
12012			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12013				    "Tag Type: %d\n", io->taskio.task_action,
12014				    io->taskio.tag_num, io->taskio.tag_type);
12015			break;
12016		default:
12017			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12018			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12019			break;
12020		}
12021		sbuf_cat(&sb, path_str);
12022		sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12023			    (intmax_t)time_uptime - io->io_hdr.start_time);
12024		sbuf_finish(&sb);
12025		printf("%s", sbuf_data(&sb));
12026	}
12027#endif /* CTL_TIME_IO */
12028
12029	mtx_lock(&control_softc->ctl_lock);
12030#ifdef CTL_IO_DELAY
12031	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12032		struct ctl_lun *lun;
12033
12034		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12035
12036		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12037	} else {
12038		struct ctl_lun *lun;
12039
12040		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12041		if ((lun != NULL)
12042		 && (lun->delay_info.datamove_delay > 0)) {
12043			struct callout *callout;
12044
12045			callout = (struct callout *)&io->io_hdr.timer_bytes;
12046			callout_init(callout, /*mpsafe*/ 1);
12047			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12048			callout_reset(callout,
12049				      lun->delay_info.datamove_delay * hz,
12050				      ctl_datamove_timer_wakeup, io);
12051			if (lun->delay_info.datamove_type ==
12052			    CTL_DELAY_TYPE_ONESHOT)
12053				lun->delay_info.datamove_delay = 0;
12054			mtx_unlock(&control_softc->ctl_lock);
12055			return;
12056		}
12057	}
12058#endif
12059	/*
12060	 * If we have any pending task management commands, process them
12061	 * first.  This is necessary to eliminate a race condition with the
12062	 * FETD:
12063	 *
12064	 * - FETD submits a task management command, like an abort.
12065	 * - Back end calls fe_datamove() to move the data for the aborted
12066	 *   command.  The FETD can't really accept it, but if it did, it
12067	 *   would end up transmitting data for a command that the initiator
12068	 *   told us to abort.
12069	 *
12070	 * We close the race by processing all pending task management
12071	 * commands here (we can't block!), and then check this I/O to see
12072	 * if it has been aborted.  If so, return it to the back end with
12073	 * bad status, so the back end can say return an error to the back end
12074	 * and then when the back end returns an error, we can return the
12075	 * aborted command to the FETD, so it can clean up its resources.
12076	 */
12077	if (control_softc->flags & CTL_FLAG_TASK_PENDING)
12078		ctl_run_task_queue(control_softc);
12079
12080	/*
12081	 * This command has been aborted.  Set the port status, so we fail
12082	 * the data move.
12083	 */
12084	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12085		printf("ctl_datamove: tag 0x%04x on (%ju:%d:%ju:%d) aborted\n",
12086		       io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id,
12087		       io->io_hdr.nexus.targ_port,
12088		       (uintmax_t)io->io_hdr.nexus.targ_target.id,
12089		       io->io_hdr.nexus.targ_lun);
12090		io->io_hdr.status = CTL_CMD_ABORTED;
12091		io->io_hdr.port_status = 31337;
12092		mtx_unlock(&control_softc->ctl_lock);
12093		/*
12094		 * Note that the backend, in this case, will get the
12095		 * callback in its context.  In other cases it may get
12096		 * called in the frontend's interrupt thread context.
12097		 */
12098		io->scsiio.be_move_done(io);
12099		return;
12100	}
12101
12102	/*
12103	 * If we're in XFER mode and this I/O is from the other shelf
12104	 * controller, we need to send the DMA to the other side to
12105	 * actually transfer the data to/from the host.  In serialize only
12106	 * mode the transfer happens below CTL and ctl_datamove() is only
12107	 * called on the machine that originally received the I/O.
12108	 */
12109	if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12110	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12111		union ctl_ha_msg msg;
12112		uint32_t sg_entries_sent;
12113		int do_sg_copy;
12114		int i;
12115
12116		memset(&msg, 0, sizeof(msg));
12117		msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12118		msg.hdr.original_sc = io->io_hdr.original_sc;
12119		msg.hdr.serializing_sc = io;
12120		msg.hdr.nexus = io->io_hdr.nexus;
12121		msg.dt.flags = io->io_hdr.flags;
12122		/*
12123		 * We convert everything into a S/G list here.  We can't
12124		 * pass by reference, only by value between controllers.
12125		 * So we can't pass a pointer to the S/G list, only as many
12126		 * S/G entries as we can fit in here.  If it's possible for
12127		 * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12128		 * then we need to break this up into multiple transfers.
12129		 */
12130		if (io->scsiio.kern_sg_entries == 0) {
12131			msg.dt.kern_sg_entries = 1;
12132			/*
12133			 * If this is in cached memory, flush the cache
12134			 * before we send the DMA request to the other
12135			 * controller.  We want to do this in either the
12136			 * read or the write case.  The read case is
12137			 * straightforward.  In the write case, we want to
12138			 * make sure nothing is in the local cache that
12139			 * could overwrite the DMAed data.
12140			 */
12141			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12142				/*
12143				 * XXX KDM use bus_dmamap_sync() here.
12144				 */
12145			}
12146
12147			/*
12148			 * Convert to a physical address if this is a
12149			 * virtual address.
12150			 */
12151			if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12152				msg.dt.sg_list[0].addr =
12153					io->scsiio.kern_data_ptr;
12154			} else {
12155				/*
12156				 * XXX KDM use busdma here!
12157				 */
12158#if 0
12159				msg.dt.sg_list[0].addr = (void *)
12160					vtophys(io->scsiio.kern_data_ptr);
12161#endif
12162			}
12163
12164			msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12165			do_sg_copy = 0;
12166		} else {
12167			struct ctl_sg_entry *sgl;
12168
12169			do_sg_copy = 1;
12170			msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12171			sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
12172			if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12173				/*
12174				 * XXX KDM use bus_dmamap_sync() here.
12175				 */
12176			}
12177		}
12178
12179		msg.dt.kern_data_len = io->scsiio.kern_data_len;
12180		msg.dt.kern_total_len = io->scsiio.kern_total_len;
12181		msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12182		msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12183		msg.dt.sg_sequence = 0;
12184
12185		/*
12186		 * Loop until we've sent all of the S/G entries.  On the
12187		 * other end, we'll recompose these S/G entries into one
12188		 * contiguous list before passing it to the
12189		 */
12190		for (sg_entries_sent = 0; sg_entries_sent <
12191		     msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12192			msg.dt.cur_sg_entries = ctl_min((sizeof(msg.dt.sg_list)/
12193				sizeof(msg.dt.sg_list[0])),
12194				msg.dt.kern_sg_entries - sg_entries_sent);
12195
12196			if (do_sg_copy != 0) {
12197				struct ctl_sg_entry *sgl;
12198				int j;
12199
12200				sgl = (struct ctl_sg_entry *)
12201					io->scsiio.kern_data_ptr;
12202				/*
12203				 * If this is in cached memory, flush the cache
12204				 * before we send the DMA request to the other
12205				 * controller.  We want to do this in either
12206				 * the * read or the write case.  The read
12207				 * case is straightforward.  In the write
12208				 * case, we want to make sure nothing is
12209				 * in the local cache that could overwrite
12210				 * the DMAed data.
12211				 */
12212
12213				for (i = sg_entries_sent, j = 0;
12214				     i < msg.dt.cur_sg_entries; i++, j++) {
12215					if ((io->io_hdr.flags &
12216					     CTL_FLAG_NO_DATASYNC) == 0) {
12217						/*
12218						 * XXX KDM use bus_dmamap_sync()
12219						 */
12220					}
12221					if ((io->io_hdr.flags &
12222					     CTL_FLAG_BUS_ADDR) == 0) {
12223						/*
12224						 * XXX KDM use busdma.
12225						 */
12226#if 0
12227						msg.dt.sg_list[j].addr =(void *)
12228						       vtophys(sgl[i].addr);
12229#endif
12230					} else {
12231						msg.dt.sg_list[j].addr =
12232							sgl[i].addr;
12233					}
12234					msg.dt.sg_list[j].len = sgl[i].len;
12235				}
12236			}
12237
12238			sg_entries_sent += msg.dt.cur_sg_entries;
12239			if (sg_entries_sent >= msg.dt.kern_sg_entries)
12240				msg.dt.sg_last = 1;
12241			else
12242				msg.dt.sg_last = 0;
12243
12244			/*
12245			 * XXX KDM drop and reacquire the lock here?
12246			 */
12247			if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12248			    sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
12249				/*
12250				 * XXX do something here.
12251				 */
12252			}
12253
12254			msg.dt.sent_sg_entries = sg_entries_sent;
12255		}
12256		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12257		if (io->io_hdr.flags & CTL_FLAG_FAILOVER)
12258			ctl_failover_io(io, /*have_lock*/ 1);
12259
12260	} else {
12261
12262		/*
12263		 * Lookup the fe_datamove() function for this particular
12264		 * front end.
12265		 */
12266		fe_datamove =
12267		    control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12268		mtx_unlock(&control_softc->ctl_lock);
12269
12270		fe_datamove(io);
12271	}
12272}
12273
12274static void
12275ctl_send_datamove_done(union ctl_io *io, int have_lock)
12276{
12277	union ctl_ha_msg msg;
12278	int isc_status;
12279
12280	memset(&msg, 0, sizeof(msg));
12281
12282	msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12283	msg.hdr.original_sc = io;
12284	msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12285	msg.hdr.nexus = io->io_hdr.nexus;
12286	msg.hdr.status = io->io_hdr.status;
12287	msg.scsi.tag_num = io->scsiio.tag_num;
12288	msg.scsi.tag_type = io->scsiio.tag_type;
12289	msg.scsi.scsi_status = io->scsiio.scsi_status;
12290	memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12291	       sizeof(io->scsiio.sense_data));
12292	msg.scsi.sense_len = io->scsiio.sense_len;
12293	msg.scsi.sense_residual = io->scsiio.sense_residual;
12294	msg.scsi.fetd_status = io->io_hdr.port_status;
12295	msg.scsi.residual = io->scsiio.residual;
12296	io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12297
12298	if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12299		ctl_failover_io(io, /*have_lock*/ have_lock);
12300		return;
12301	}
12302
12303	isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0);
12304	if (isc_status > CTL_HA_STATUS_SUCCESS) {
12305		/* XXX do something if this fails */
12306	}
12307
12308}
12309
12310/*
12311 * The DMA to the remote side is done, now we need to tell the other side
12312 * we're done so it can continue with its data movement.
12313 */
12314static void
12315ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12316{
12317	union ctl_io *io;
12318
12319	io = rq->context;
12320
12321	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12322		printf("%s: ISC DMA write failed with error %d", __func__,
12323		       rq->ret);
12324		ctl_set_internal_failure(&io->scsiio,
12325					 /*sks_valid*/ 1,
12326					 /*retry_count*/ rq->ret);
12327	}
12328
12329	ctl_dt_req_free(rq);
12330
12331	/*
12332	 * In this case, we had to malloc the memory locally.  Free it.
12333	 */
12334	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12335		int i;
12336		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12337			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12338	}
12339	/*
12340	 * The data is in local and remote memory, so now we need to send
12341	 * status (good or back) back to the other side.
12342	 */
12343	ctl_send_datamove_done(io, /*have_lock*/ 0);
12344}
12345
12346/*
12347 * We've moved the data from the host/controller into local memory.  Now we
12348 * need to push it over to the remote controller's memory.
12349 */
12350static int
12351ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12352{
12353	int retval;
12354
12355	retval = 0;
12356
12357	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12358					  ctl_datamove_remote_write_cb);
12359
12360	return (retval);
12361}
12362
12363static void
12364ctl_datamove_remote_write(union ctl_io *io)
12365{
12366	int retval;
12367	void (*fe_datamove)(union ctl_io *io);
12368
12369	/*
12370	 * - Get the data from the host/HBA into local memory.
12371	 * - DMA memory from the local controller to the remote controller.
12372	 * - Send status back to the remote controller.
12373	 */
12374
12375	retval = ctl_datamove_remote_sgl_setup(io);
12376	if (retval != 0)
12377		return;
12378
12379	/* Switch the pointer over so the FETD knows what to do */
12380	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12381
12382	/*
12383	 * Use a custom move done callback, since we need to send completion
12384	 * back to the other controller, not to the backend on this side.
12385	 */
12386	io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12387
12388	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12389
12390	fe_datamove(io);
12391
12392	return;
12393
12394}
12395
12396static int
12397ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12398{
12399#if 0
12400	char str[256];
12401	char path_str[64];
12402	struct sbuf sb;
12403#endif
12404
12405	/*
12406	 * In this case, we had to malloc the memory locally.  Free it.
12407	 */
12408	if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) {
12409		int i;
12410		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12411			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12412	}
12413
12414#if 0
12415	scsi_path_string(io, path_str, sizeof(path_str));
12416	sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12417	sbuf_cat(&sb, path_str);
12418	scsi_command_string(&io->scsiio, NULL, &sb);
12419	sbuf_printf(&sb, "\n");
12420	sbuf_cat(&sb, path_str);
12421	sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12422		    io->scsiio.tag_num, io->scsiio.tag_type);
12423	sbuf_cat(&sb, path_str);
12424	sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12425		    io->io_hdr.flags, io->io_hdr.status);
12426	sbuf_finish(&sb);
12427	printk("%s", sbuf_data(&sb));
12428#endif
12429
12430
12431	/*
12432	 * The read is done, now we need to send status (good or bad) back
12433	 * to the other side.
12434	 */
12435	ctl_send_datamove_done(io, /*have_lock*/ 0);
12436
12437	return (0);
12438}
12439
12440static void
12441ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12442{
12443	union ctl_io *io;
12444	void (*fe_datamove)(union ctl_io *io);
12445
12446	io = rq->context;
12447
12448	if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12449		printf("%s: ISC DMA read failed with error %d", __func__,
12450		       rq->ret);
12451		ctl_set_internal_failure(&io->scsiio,
12452					 /*sks_valid*/ 1,
12453					 /*retry_count*/ rq->ret);
12454	}
12455
12456	ctl_dt_req_free(rq);
12457
12458	/* Switch the pointer over so the FETD knows what to do */
12459	io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12460
12461	/*
12462	 * Use a custom move done callback, since we need to send completion
12463	 * back to the other controller, not to the backend on this side.
12464	 */
12465	io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12466
12467	/* XXX KDM add checks like the ones in ctl_datamove? */
12468
12469	fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove;
12470
12471	fe_datamove(io);
12472}
12473
12474static int
12475ctl_datamove_remote_sgl_setup(union ctl_io *io)
12476{
12477	struct ctl_sg_entry *local_sglist, *remote_sglist;
12478	struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist;
12479	struct ctl_softc *softc;
12480	int retval;
12481	int i;
12482
12483	retval = 0;
12484	softc = control_softc;
12485
12486	local_sglist = io->io_hdr.local_sglist;
12487	local_dma_sglist = io->io_hdr.local_dma_sglist;
12488	remote_sglist = io->io_hdr.remote_sglist;
12489	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12490
12491	if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) {
12492		for (i = 0; i < io->scsiio.kern_sg_entries; i++) {
12493			local_sglist[i].len = remote_sglist[i].len;
12494
12495			/*
12496			 * XXX Detect the situation where the RS-level I/O
12497			 * redirector on the other side has already read the
12498			 * data off of the AOR RS on this side, and
12499			 * transferred it to remote (mirror) memory on the
12500			 * other side.  Since we already have the data in
12501			 * memory here, we just need to use it.
12502			 *
12503			 * XXX KDM this can probably be removed once we
12504			 * get the cache device code in and take the
12505			 * current AOR implementation out.
12506			 */
12507#ifdef NEEDTOPORT
12508			if ((remote_sglist[i].addr >=
12509			     (void *)vtophys(softc->mirr->addr))
12510			 && (remote_sglist[i].addr <
12511			     ((void *)vtophys(softc->mirr->addr) +
12512			     CacheMirrorOffset))) {
12513				local_sglist[i].addr = remote_sglist[i].addr -
12514					CacheMirrorOffset;
12515				if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12516				     CTL_FLAG_DATA_IN)
12517					io->io_hdr.flags |= CTL_FLAG_REDIR_DONE;
12518			} else {
12519				local_sglist[i].addr = remote_sglist[i].addr +
12520					CacheMirrorOffset;
12521			}
12522#endif
12523#if 0
12524			printf("%s: local %p, remote %p, len %d\n",
12525			       __func__, local_sglist[i].addr,
12526			       remote_sglist[i].addr, local_sglist[i].len);
12527#endif
12528		}
12529	} else {
12530		uint32_t len_to_go;
12531
12532		/*
12533		 * In this case, we don't have automatically allocated
12534		 * memory for this I/O on this controller.  This typically
12535		 * happens with internal CTL I/O -- e.g. inquiry, mode
12536		 * sense, etc.  Anything coming from RAIDCore will have
12537		 * a mirror area available.
12538		 */
12539		len_to_go = io->scsiio.kern_data_len;
12540
12541		/*
12542		 * Clear the no datasync flag, we have to use malloced
12543		 * buffers.
12544		 */
12545		io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC;
12546
12547		/*
12548		 * The difficult thing here is that the size of the various
12549		 * S/G segments may be different than the size from the
12550		 * remote controller.  That'll make it harder when DMAing
12551		 * the data back to the other side.
12552		 */
12553		for (i = 0; (i < sizeof(io->io_hdr.remote_sglist) /
12554		     sizeof(io->io_hdr.remote_sglist[0])) &&
12555		     (len_to_go > 0); i++) {
12556			local_sglist[i].len = ctl_min(len_to_go, 131072);
12557			CTL_SIZE_8B(local_dma_sglist[i].len,
12558				    local_sglist[i].len);
12559			local_sglist[i].addr =
12560				malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK);
12561
12562			local_dma_sglist[i].addr = local_sglist[i].addr;
12563
12564			if (local_sglist[i].addr == NULL) {
12565				int j;
12566
12567				printf("malloc failed for %zd bytes!",
12568				       local_dma_sglist[i].len);
12569				for (j = 0; j < i; j++) {
12570					free(local_sglist[j].addr, M_CTL);
12571				}
12572				ctl_set_internal_failure(&io->scsiio,
12573							 /*sks_valid*/ 1,
12574							 /*retry_count*/ 4857);
12575				retval = 1;
12576				goto bailout_error;
12577
12578			}
12579			/* XXX KDM do we need a sync here? */
12580
12581			len_to_go -= local_sglist[i].len;
12582		}
12583		/*
12584		 * Reset the number of S/G entries accordingly.  The
12585		 * original number of S/G entries is available in
12586		 * rem_sg_entries.
12587		 */
12588		io->scsiio.kern_sg_entries = i;
12589
12590#if 0
12591		printf("%s: kern_sg_entries = %d\n", __func__,
12592		       io->scsiio.kern_sg_entries);
12593		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12594			printf("%s: sg[%d] = %p, %d (DMA: %d)\n", __func__, i,
12595			       local_sglist[i].addr, local_sglist[i].len,
12596			       local_dma_sglist[i].len);
12597#endif
12598	}
12599
12600
12601	return (retval);
12602
12603bailout_error:
12604
12605	ctl_send_datamove_done(io, /*have_lock*/ 0);
12606
12607	return (retval);
12608}
12609
12610static int
12611ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12612			 ctl_ha_dt_cb callback)
12613{
12614	struct ctl_ha_dt_req *rq;
12615	struct ctl_sg_entry *remote_sglist, *local_sglist;
12616	struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist;
12617	uint32_t local_used, remote_used, total_used;
12618	int retval;
12619	int i, j;
12620
12621	retval = 0;
12622
12623	rq = ctl_dt_req_alloc();
12624
12625	/*
12626	 * If we failed to allocate the request, and if the DMA didn't fail
12627	 * anyway, set busy status.  This is just a resource allocation
12628	 * failure.
12629	 */
12630	if ((rq == NULL)
12631	 && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12632		ctl_set_busy(&io->scsiio);
12633
12634	if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12635
12636		if (rq != NULL)
12637			ctl_dt_req_free(rq);
12638
12639		/*
12640		 * The data move failed.  We need to return status back
12641		 * to the other controller.  No point in trying to DMA
12642		 * data to the remote controller.
12643		 */
12644
12645		ctl_send_datamove_done(io, /*have_lock*/ 0);
12646
12647		retval = 1;
12648
12649		goto bailout;
12650	}
12651
12652	local_sglist = io->io_hdr.local_sglist;
12653	local_dma_sglist = io->io_hdr.local_dma_sglist;
12654	remote_sglist = io->io_hdr.remote_sglist;
12655	remote_dma_sglist = io->io_hdr.remote_dma_sglist;
12656	local_used = 0;
12657	remote_used = 0;
12658	total_used = 0;
12659
12660	if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) {
12661		rq->ret = CTL_HA_STATUS_SUCCESS;
12662		rq->context = io;
12663		callback(rq);
12664		goto bailout;
12665	}
12666
12667	/*
12668	 * Pull/push the data over the wire from/to the other controller.
12669	 * This takes into account the possibility that the local and
12670	 * remote sglists may not be identical in terms of the size of
12671	 * the elements and the number of elements.
12672	 *
12673	 * One fundamental assumption here is that the length allocated for
12674	 * both the local and remote sglists is identical.  Otherwise, we've
12675	 * essentially got a coding error of some sort.
12676	 */
12677	for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12678		int isc_ret;
12679		uint32_t cur_len, dma_length;
12680		uint8_t *tmp_ptr;
12681
12682		rq->id = CTL_HA_DATA_CTL;
12683		rq->command = command;
12684		rq->context = io;
12685
12686		/*
12687		 * Both pointers should be aligned.  But it is possible
12688		 * that the allocation length is not.  They should both
12689		 * also have enough slack left over at the end, though,
12690		 * to round up to the next 8 byte boundary.
12691		 */
12692		cur_len = ctl_min(local_sglist[i].len - local_used,
12693				  remote_sglist[j].len - remote_used);
12694
12695		/*
12696		 * In this case, we have a size issue and need to decrease
12697		 * the size, except in the case where we actually have less
12698		 * than 8 bytes left.  In that case, we need to increase
12699		 * the DMA length to get the last bit.
12700		 */
12701		if ((cur_len & 0x7) != 0) {
12702			if (cur_len > 0x7) {
12703				cur_len = cur_len - (cur_len & 0x7);
12704				dma_length = cur_len;
12705			} else {
12706				CTL_SIZE_8B(dma_length, cur_len);
12707			}
12708
12709		} else
12710			dma_length = cur_len;
12711
12712		/*
12713		 * If we had to allocate memory for this I/O, instead of using
12714		 * the non-cached mirror memory, we'll need to flush the cache
12715		 * before trying to DMA to the other controller.
12716		 *
12717		 * We could end up doing this multiple times for the same
12718		 * segment if we have a larger local segment than remote
12719		 * segment.  That shouldn't be an issue.
12720		 */
12721		if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) {
12722			/*
12723			 * XXX KDM use bus_dmamap_sync() here.
12724			 */
12725		}
12726
12727		rq->size = dma_length;
12728
12729		tmp_ptr = (uint8_t *)local_sglist[i].addr;
12730		tmp_ptr += local_used;
12731
12732		/* Use physical addresses when talking to ISC hardware */
12733		if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12734			/* XXX KDM use busdma */
12735#if 0
12736			rq->local = vtophys(tmp_ptr);
12737#endif
12738		} else
12739			rq->local = tmp_ptr;
12740
12741		tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12742		tmp_ptr += remote_used;
12743		rq->remote = tmp_ptr;
12744
12745		rq->callback = NULL;
12746
12747		local_used += cur_len;
12748		if (local_used >= local_sglist[i].len) {
12749			i++;
12750			local_used = 0;
12751		}
12752
12753		remote_used += cur_len;
12754		if (remote_used >= remote_sglist[j].len) {
12755			j++;
12756			remote_used = 0;
12757		}
12758		total_used += cur_len;
12759
12760		if (total_used >= io->scsiio.kern_data_len)
12761			rq->callback = callback;
12762
12763		if ((rq->size & 0x7) != 0) {
12764			printf("%s: warning: size %d is not on 8b boundary\n",
12765			       __func__, rq->size);
12766		}
12767		if (((uintptr_t)rq->local & 0x7) != 0) {
12768			printf("%s: warning: local %p not on 8b boundary\n",
12769			       __func__, rq->local);
12770		}
12771		if (((uintptr_t)rq->remote & 0x7) != 0) {
12772			printf("%s: warning: remote %p not on 8b boundary\n",
12773			       __func__, rq->local);
12774		}
12775#if 0
12776		printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12777		       (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12778		       rq->local, rq->remote, rq->size);
12779#endif
12780
12781		isc_ret = ctl_dt_single(rq);
12782		if (isc_ret == CTL_HA_STATUS_WAIT)
12783			continue;
12784
12785		if (isc_ret == CTL_HA_STATUS_DISCONNECT) {
12786			rq->ret = CTL_HA_STATUS_SUCCESS;
12787		} else {
12788			rq->ret = isc_ret;
12789		}
12790		callback(rq);
12791		goto bailout;
12792	}
12793
12794bailout:
12795	return (retval);
12796
12797}
12798
12799static void
12800ctl_datamove_remote_read(union ctl_io *io)
12801{
12802	int retval;
12803	int i;
12804
12805	/*
12806	 * This will send an error to the other controller in the case of a
12807	 * failure.
12808	 */
12809	retval = ctl_datamove_remote_sgl_setup(io);
12810	if (retval != 0)
12811		return;
12812
12813	retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12814					  ctl_datamove_remote_read_cb);
12815	if ((retval != 0)
12816	 && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0)) {
12817		/*
12818		 * Make sure we free memory if there was an error..  The
12819		 * ctl_datamove_remote_xfer() function will send the
12820		 * datamove done message, or call the callback with an
12821		 * error if there is a problem.
12822		 */
12823		for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12824			free(io->io_hdr.local_sglist[i].addr, M_CTL);
12825	}
12826
12827	return;
12828}
12829
12830/*
12831 * Process a datamove request from the other controller.  This is used for
12832 * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12833 * first.  Once that is complete, the data gets DMAed into the remote
12834 * controller's memory.  For reads, we DMA from the remote controller's
12835 * memory into our memory first, and then move it out to the FETD.
12836 */
12837static void
12838ctl_datamove_remote(union ctl_io *io)
12839{
12840	struct ctl_softc *softc;
12841
12842	softc = control_softc;
12843
12844	mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
12845
12846	/*
12847	 * Note that we look for an aborted I/O here, but don't do some of
12848	 * the other checks that ctl_datamove() normally does.  We don't
12849	 * need to run the task queue, because this I/O is on the ISC
12850	 * queue, which is executed by the work thread after the task queue.
12851	 * We don't need to run the datamove delay code, since that should
12852	 * have been done if need be on the other controller.
12853	 */
12854	mtx_lock(&softc->ctl_lock);
12855
12856	if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12857
12858		printf("%s: tag 0x%04x on (%d:%d:%d:%d) aborted\n", __func__,
12859		       io->scsiio.tag_num, io->io_hdr.nexus.initid.id,
12860		       io->io_hdr.nexus.targ_port,
12861		       io->io_hdr.nexus.targ_target.id,
12862		       io->io_hdr.nexus.targ_lun);
12863		io->io_hdr.status = CTL_CMD_ABORTED;
12864		io->io_hdr.port_status = 31338;
12865
12866		mtx_unlock(&softc->ctl_lock);
12867
12868		ctl_send_datamove_done(io, /*have_lock*/ 0);
12869
12870		return;
12871	}
12872
12873	if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) {
12874		mtx_unlock(&softc->ctl_lock);
12875		ctl_datamove_remote_write(io);
12876	} else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN){
12877		mtx_unlock(&softc->ctl_lock);
12878		ctl_datamove_remote_read(io);
12879	} else {
12880		union ctl_ha_msg msg;
12881		struct scsi_sense_data *sense;
12882		uint8_t sks[3];
12883		int retry_count;
12884
12885		memset(&msg, 0, sizeof(msg));
12886
12887		msg.hdr.msg_type = CTL_MSG_BAD_JUJU;
12888		msg.hdr.status = CTL_SCSI_ERROR;
12889		msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND;
12890
12891		retry_count = 4243;
12892
12893		sense = &msg.scsi.sense_data;
12894		sks[0] = SSD_SCS_VALID;
12895		sks[1] = (retry_count >> 8) & 0xff;
12896		sks[2] = retry_count & 0xff;
12897
12898		/* "Internal target failure" */
12899		scsi_set_sense_data(sense,
12900				    /*sense_format*/ SSD_TYPE_NONE,
12901				    /*current_error*/ 1,
12902				    /*sense_key*/ SSD_KEY_HARDWARE_ERROR,
12903				    /*asc*/ 0x44,
12904				    /*ascq*/ 0x00,
12905				    /*type*/ SSD_ELEM_SKS,
12906				    /*size*/ sizeof(sks),
12907				    /*data*/ sks,
12908				    SSD_ELEM_NONE);
12909
12910		io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12911		if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12912			ctl_failover_io(io, /*have_lock*/ 1);
12913			mtx_unlock(&softc->ctl_lock);
12914			return;
12915		}
12916
12917		mtx_unlock(&softc->ctl_lock);
12918
12919		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) >
12920		    CTL_HA_STATUS_SUCCESS) {
12921			/* XXX KDM what to do if this fails? */
12922		}
12923		return;
12924	}
12925
12926}
12927
12928static int
12929ctl_process_done(union ctl_io *io, int have_lock)
12930{
12931	struct ctl_lun *lun;
12932	struct ctl_softc *ctl_softc;
12933	void (*fe_done)(union ctl_io *io);
12934	uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port);
12935
12936	CTL_DEBUG_PRINT(("ctl_process_done\n"));
12937
12938	fe_done =
12939	    control_softc->ctl_ports[targ_port]->fe_done;
12940
12941#ifdef CTL_TIME_IO
12942	if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12943		char str[256];
12944		char path_str[64];
12945		struct sbuf sb;
12946
12947		ctl_scsi_path_string(io, path_str, sizeof(path_str));
12948		sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12949
12950		sbuf_cat(&sb, path_str);
12951		switch (io->io_hdr.io_type) {
12952		case CTL_IO_SCSI:
12953			ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12954			sbuf_printf(&sb, "\n");
12955			sbuf_cat(&sb, path_str);
12956			sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12957				    io->scsiio.tag_num, io->scsiio.tag_type);
12958			break;
12959		case CTL_IO_TASK:
12960			sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12961				    "Tag Type: %d\n", io->taskio.task_action,
12962				    io->taskio.tag_num, io->taskio.tag_type);
12963			break;
12964		default:
12965			printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12966			panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12967			break;
12968		}
12969		sbuf_cat(&sb, path_str);
12970		sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12971			    (intmax_t)time_uptime - io->io_hdr.start_time);
12972		sbuf_finish(&sb);
12973		printf("%s", sbuf_data(&sb));
12974	}
12975#endif /* CTL_TIME_IO */
12976
12977	switch (io->io_hdr.io_type) {
12978	case CTL_IO_SCSI:
12979		break;
12980	case CTL_IO_TASK:
12981		if (bootverbose || verbose > 0)
12982			ctl_io_error_print(io, NULL);
12983		if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12984			ctl_free_io(io);
12985		else
12986			fe_done(io);
12987		return (CTL_RETVAL_COMPLETE);
12988		break;
12989	default:
12990		printf("ctl_process_done: invalid io type %d\n",
12991		       io->io_hdr.io_type);
12992		panic("ctl_process_done: invalid io type %d\n",
12993		      io->io_hdr.io_type);
12994		break; /* NOTREACHED */
12995	}
12996
12997	lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12998	if (lun == NULL) {
12999		CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13000				 io->io_hdr.nexus.targ_lun));
13001		fe_done(io);
13002		goto bailout;
13003	}
13004	ctl_softc = lun->ctl_softc;
13005
13006	/*
13007	 * Remove this from the OOA queue.
13008	 */
13009	if (have_lock == 0)
13010		mtx_lock(&ctl_softc->ctl_lock);
13011
13012	/*
13013	 * Check to see if we have any errors to inject here.  We only
13014	 * inject errors for commands that don't already have errors set.
13015	 */
13016	if ((STAILQ_FIRST(&lun->error_list) != NULL)
13017	 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS))
13018		ctl_inject_error(lun, io);
13019
13020	/*
13021	 * XXX KDM how do we treat commands that aren't completed
13022	 * successfully?
13023	 *
13024	 * XXX KDM should we also track I/O latency?
13025	 */
13026	if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) {
13027		uint32_t blocksize;
13028#ifdef CTL_TIME_IO
13029		struct bintime cur_bt;
13030#endif
13031
13032		if ((lun->be_lun != NULL)
13033		 && (lun->be_lun->blocksize != 0))
13034			blocksize = lun->be_lun->blocksize;
13035		else
13036			blocksize = 512;
13037
13038		switch (io->io_hdr.io_type) {
13039		case CTL_IO_SCSI: {
13040			int isread;
13041			struct ctl_lba_len_flags *lbalen;
13042
13043			isread = 0;
13044			switch (io->scsiio.cdb[0]) {
13045			case READ_6:
13046			case READ_10:
13047			case READ_12:
13048			case READ_16:
13049				isread = 1;
13050				/* FALLTHROUGH */
13051			case WRITE_6:
13052			case WRITE_10:
13053			case WRITE_12:
13054			case WRITE_16:
13055			case WRITE_VERIFY_10:
13056			case WRITE_VERIFY_12:
13057			case WRITE_VERIFY_16:
13058				lbalen = (struct ctl_lba_len_flags *)
13059				    &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
13060
13061				if (isread) {
13062					lun->stats.ports[targ_port].bytes[CTL_STATS_READ] +=
13063					    lbalen->len * blocksize;
13064					lun->stats.ports[targ_port].operations[CTL_STATS_READ]++;
13065
13066#ifdef CTL_TIME_IO
13067					bintime_add(
13068					   &lun->stats.ports[targ_port].dma_time[CTL_STATS_READ],
13069					   &io->io_hdr.dma_bt);
13070					lun->stats.ports[targ_port].num_dmas[CTL_STATS_READ] +=
13071						io->io_hdr.num_dmas;
13072					getbintime(&cur_bt);
13073					bintime_sub(&cur_bt,
13074						    &io->io_hdr.start_bt);
13075
13076					bintime_add(
13077					    &lun->stats.ports[targ_port].time[CTL_STATS_READ],
13078					    &cur_bt);
13079
13080#if 0
13081					cs_prof_gettime(&cur_ticks);
13082					lun->stats.time[CTL_STATS_READ] +=
13083						cur_ticks -
13084						io->io_hdr.start_ticks;
13085#endif
13086#if 0
13087					lun->stats.time[CTL_STATS_READ] +=
13088						jiffies - io->io_hdr.start_time;
13089#endif
13090#endif /* CTL_TIME_IO */
13091				} else {
13092					lun->stats.ports[targ_port].bytes[CTL_STATS_WRITE] +=
13093					    lbalen->len * blocksize;
13094					lun->stats.ports[targ_port].operations[
13095						CTL_STATS_WRITE]++;
13096
13097#ifdef CTL_TIME_IO
13098					bintime_add(
13099					  &lun->stats.ports[targ_port].dma_time[CTL_STATS_WRITE],
13100					  &io->io_hdr.dma_bt);
13101					lun->stats.ports[targ_port].num_dmas[CTL_STATS_WRITE] +=
13102						io->io_hdr.num_dmas;
13103					getbintime(&cur_bt);
13104					bintime_sub(&cur_bt,
13105						    &io->io_hdr.start_bt);
13106
13107					bintime_add(
13108					    &lun->stats.ports[targ_port].time[CTL_STATS_WRITE],
13109					    &cur_bt);
13110#if 0
13111					cs_prof_gettime(&cur_ticks);
13112					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
13113						cur_ticks -
13114						io->io_hdr.start_ticks;
13115					lun->stats.ports[targ_port].time[CTL_STATS_WRITE] +=
13116						jiffies - io->io_hdr.start_time;
13117#endif
13118#endif /* CTL_TIME_IO */
13119				}
13120				break;
13121			default:
13122				lun->stats.ports[targ_port].operations[CTL_STATS_NO_IO]++;
13123
13124#ifdef CTL_TIME_IO
13125				bintime_add(
13126				  &lun->stats.ports[targ_port].dma_time[CTL_STATS_NO_IO],
13127				  &io->io_hdr.dma_bt);
13128				lun->stats.ports[targ_port].num_dmas[CTL_STATS_NO_IO] +=
13129					io->io_hdr.num_dmas;
13130				getbintime(&cur_bt);
13131				bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13132
13133				bintime_add(&lun->stats.ports[targ_port].time[CTL_STATS_NO_IO],
13134					    &cur_bt);
13135
13136#if 0
13137				cs_prof_gettime(&cur_ticks);
13138				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
13139					cur_ticks -
13140					io->io_hdr.start_ticks;
13141				lun->stats.ports[targ_port].time[CTL_STATS_NO_IO] +=
13142					jiffies - io->io_hdr.start_time;
13143#endif
13144#endif /* CTL_TIME_IO */
13145				break;
13146			}
13147			break;
13148		}
13149		default:
13150			break;
13151		}
13152	}
13153
13154	TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13155
13156	/*
13157	 * Run through the blocked queue on this LUN and see if anything
13158	 * has become unblocked, now that this transaction is done.
13159	 */
13160	ctl_check_blocked(lun);
13161
13162	/*
13163	 * If the LUN has been invalidated, free it if there is nothing
13164	 * left on its OOA queue.
13165	 */
13166	if ((lun->flags & CTL_LUN_INVALID)
13167	 && (TAILQ_FIRST(&lun->ooa_queue) == NULL))
13168		ctl_free_lun(lun);
13169
13170	/*
13171	 * If this command has been aborted, make sure we set the status
13172	 * properly.  The FETD is responsible for freeing the I/O and doing
13173	 * whatever it needs to do to clean up its state.
13174	 */
13175	if (io->io_hdr.flags & CTL_FLAG_ABORT)
13176		io->io_hdr.status = CTL_CMD_ABORTED;
13177
13178	/*
13179	 * We print out status for every task management command.  For SCSI
13180	 * commands, we filter out any unit attention errors; they happen
13181	 * on every boot, and would clutter up the log.  Note:  task
13182	 * management commands aren't printed here, they are printed above,
13183	 * since they should never even make it down here.
13184	 */
13185	switch (io->io_hdr.io_type) {
13186	case CTL_IO_SCSI: {
13187		int error_code, sense_key, asc, ascq;
13188
13189		sense_key = 0;
13190
13191		if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR)
13192		 && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) {
13193			/*
13194			 * Since this is just for printing, no need to
13195			 * show errors here.
13196			 */
13197			scsi_extract_sense_len(&io->scsiio.sense_data,
13198					       io->scsiio.sense_len,
13199					       &error_code,
13200					       &sense_key,
13201					       &asc,
13202					       &ascq,
13203					       /*show_errors*/ 0);
13204		}
13205
13206		if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)
13207		 && (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SCSI_ERROR)
13208		  || (io->scsiio.scsi_status != SCSI_STATUS_CHECK_COND)
13209		  || (sense_key != SSD_KEY_UNIT_ATTENTION))) {
13210
13211			if ((time_uptime - ctl_softc->last_print_jiffies) <= 0){
13212				ctl_softc->skipped_prints++;
13213				if (have_lock == 0)
13214					mtx_unlock(&ctl_softc->ctl_lock);
13215			} else {
13216				uint32_t skipped_prints;
13217
13218				skipped_prints = ctl_softc->skipped_prints;
13219
13220				ctl_softc->skipped_prints = 0;
13221				ctl_softc->last_print_jiffies = time_uptime;
13222
13223				if (have_lock == 0)
13224					mtx_unlock(&ctl_softc->ctl_lock);
13225				if (skipped_prints > 0) {
13226#ifdef NEEDTOPORT
13227					csevent_log(CSC_CTL | CSC_SHELF_SW |
13228					    CTL_ERROR_REPORT,
13229					    csevent_LogType_Trace,
13230					    csevent_Severity_Information,
13231					    csevent_AlertLevel_Green,
13232					    csevent_FRU_Firmware,
13233					    csevent_FRU_Unknown,
13234					    "High CTL error volume, %d prints "
13235					    "skipped", skipped_prints);
13236#endif
13237				}
13238				if (bootverbose || verbose > 0)
13239					ctl_io_error_print(io, NULL);
13240			}
13241		} else {
13242			if (have_lock == 0)
13243				mtx_unlock(&ctl_softc->ctl_lock);
13244		}
13245		break;
13246	}
13247	case CTL_IO_TASK:
13248		if (have_lock == 0)
13249			mtx_unlock(&ctl_softc->ctl_lock);
13250		if (bootverbose || verbose > 0)
13251			ctl_io_error_print(io, NULL);
13252		break;
13253	default:
13254		if (have_lock == 0)
13255			mtx_unlock(&ctl_softc->ctl_lock);
13256		break;
13257	}
13258
13259	/*
13260	 * Tell the FETD or the other shelf controller we're done with this
13261	 * command.  Note that only SCSI commands get to this point.  Task
13262	 * management commands are completed above.
13263	 *
13264	 * We only send status to the other controller if we're in XFER
13265	 * mode.  In SER_ONLY mode, the I/O is done on the controller that
13266	 * received the I/O (from CTL's perspective), and so the status is
13267	 * generated there.
13268	 *
13269	 * XXX KDM if we hold the lock here, we could cause a deadlock
13270	 * if the frontend comes back in in this context to queue
13271	 * something.
13272	 */
13273	if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER)
13274	 && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13275		union ctl_ha_msg msg;
13276
13277		memset(&msg, 0, sizeof(msg));
13278		msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13279		msg.hdr.original_sc = io->io_hdr.original_sc;
13280		msg.hdr.nexus = io->io_hdr.nexus;
13281		msg.hdr.status = io->io_hdr.status;
13282		msg.scsi.scsi_status = io->scsiio.scsi_status;
13283		msg.scsi.tag_num = io->scsiio.tag_num;
13284		msg.scsi.tag_type = io->scsiio.tag_type;
13285		msg.scsi.sense_len = io->scsiio.sense_len;
13286		msg.scsi.sense_residual = io->scsiio.sense_residual;
13287		msg.scsi.residual = io->scsiio.residual;
13288		memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13289		       sizeof(io->scsiio.sense_data));
13290		/*
13291		 * We copy this whether or not this is an I/O-related
13292		 * command.  Otherwise, we'd have to go and check to see
13293		 * whether it's a read/write command, and it really isn't
13294		 * worth it.
13295		 */
13296		memcpy(&msg.scsi.lbalen,
13297		       &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
13298		       sizeof(msg.scsi.lbalen));
13299
13300		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13301				sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) {
13302			/* XXX do something here */
13303		}
13304
13305		ctl_free_io(io);
13306	} else
13307		fe_done(io);
13308
13309bailout:
13310
13311	return (CTL_RETVAL_COMPLETE);
13312}
13313
13314/*
13315 * Front end should call this if it doesn't do autosense.  When the request
13316 * sense comes back in from the initiator, we'll dequeue this and send it.
13317 */
13318int
13319ctl_queue_sense(union ctl_io *io)
13320{
13321	struct ctl_lun *lun;
13322	struct ctl_softc *ctl_softc;
13323	uint32_t initidx, targ_lun;
13324
13325	ctl_softc = control_softc;
13326
13327	CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13328
13329	/*
13330	 * LUN lookup will likely move to the ctl_work_thread() once we
13331	 * have our new queueing infrastructure (that doesn't put things on
13332	 * a per-LUN queue initially).  That is so that we can handle
13333	 * things like an INQUIRY to a LUN that we don't have enabled.  We
13334	 * can't deal with that right now.
13335	 */
13336	mtx_lock(&ctl_softc->ctl_lock);
13337
13338	/*
13339	 * If we don't have a LUN for this, just toss the sense
13340	 * information.
13341	 */
13342	targ_lun = io->io_hdr.nexus.targ_lun;
13343	if (io->io_hdr.nexus.lun_map_fn != NULL)
13344		targ_lun = io->io_hdr.nexus.lun_map_fn(io->io_hdr.nexus.lun_map_arg, targ_lun);
13345	if ((targ_lun < CTL_MAX_LUNS)
13346	 && (ctl_softc->ctl_luns[targ_lun] != NULL))
13347		lun = ctl_softc->ctl_luns[targ_lun];
13348	else
13349		goto bailout;
13350
13351	initidx = ctl_get_initindex(&io->io_hdr.nexus);
13352
13353	/*
13354	 * Already have CA set for this LUN...toss the sense information.
13355	 */
13356	if (ctl_is_set(lun->have_ca, initidx))
13357		goto bailout;
13358
13359	memcpy(&lun->pending_sense[initidx].sense, &io->scsiio.sense_data,
13360	       ctl_min(sizeof(lun->pending_sense[initidx].sense),
13361	       sizeof(io->scsiio.sense_data)));
13362	ctl_set_mask(lun->have_ca, initidx);
13363
13364bailout:
13365	mtx_unlock(&ctl_softc->ctl_lock);
13366
13367	ctl_free_io(io);
13368
13369	return (CTL_RETVAL_COMPLETE);
13370}
13371
13372/*
13373 * Primary command inlet from frontend ports.  All SCSI and task I/O
13374 * requests must go through this function.
13375 */
13376int
13377ctl_queue(union ctl_io *io)
13378{
13379	struct ctl_softc *ctl_softc;
13380
13381	CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13382
13383	ctl_softc = control_softc;
13384
13385#ifdef CTL_TIME_IO
13386	io->io_hdr.start_time = time_uptime;
13387	getbintime(&io->io_hdr.start_bt);
13388#endif /* CTL_TIME_IO */
13389
13390	mtx_lock(&ctl_softc->ctl_lock);
13391
13392	switch (io->io_hdr.io_type) {
13393	case CTL_IO_SCSI:
13394		STAILQ_INSERT_TAIL(&ctl_softc->incoming_queue, &io->io_hdr,
13395				   links);
13396		break;
13397	case CTL_IO_TASK:
13398		STAILQ_INSERT_TAIL(&ctl_softc->task_queue, &io->io_hdr, links);
13399		/*
13400		 * Set the task pending flag.  This is necessary to close a
13401		 * race condition with the FETD:
13402		 *
13403		 * - FETD submits a task management command, like an abort.
13404		 * - Back end calls fe_datamove() to move the data for the
13405		 *   aborted command.  The FETD can't really accept it, but
13406		 *   if it did, it would end up transmitting data for a
13407		 *   command that the initiator told us to abort.
13408		 *
13409		 * We close the race condition by setting the flag here,
13410		 * and checking it in ctl_datamove(), before calling the
13411		 * FETD's fe_datamove routine.  If we've got a task
13412		 * pending, we run the task queue and then check to see
13413		 * whether our particular I/O has been aborted.
13414		 */
13415		ctl_softc->flags |= CTL_FLAG_TASK_PENDING;
13416		break;
13417	default:
13418		mtx_unlock(&ctl_softc->ctl_lock);
13419		printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13420		return (EINVAL);
13421		break; /* NOTREACHED */
13422	}
13423	mtx_unlock(&ctl_softc->ctl_lock);
13424
13425	ctl_wakeup_thread();
13426
13427	return (CTL_RETVAL_COMPLETE);
13428}
13429
13430#ifdef CTL_IO_DELAY
13431static void
13432ctl_done_timer_wakeup(void *arg)
13433{
13434	union ctl_io *io;
13435
13436	io = (union ctl_io *)arg;
13437	ctl_done_lock(io, /*have_lock*/ 0);
13438}
13439#endif /* CTL_IO_DELAY */
13440
13441void
13442ctl_done_lock(union ctl_io *io, int have_lock)
13443{
13444	struct ctl_softc *ctl_softc;
13445#ifndef CTL_DONE_THREAD
13446	union ctl_io *xio;
13447#endif /* !CTL_DONE_THREAD */
13448
13449	ctl_softc = control_softc;
13450
13451	if (have_lock == 0)
13452		mtx_lock(&ctl_softc->ctl_lock);
13453
13454	/*
13455	 * Enable this to catch duplicate completion issues.
13456	 */
13457#if 0
13458	if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13459		printf("%s: type %d msg %d cdb %x iptl: "
13460		       "%d:%d:%d:%d tag 0x%04x "
13461		       "flag %#x status %x\n",
13462			__func__,
13463			io->io_hdr.io_type,
13464			io->io_hdr.msg_type,
13465			io->scsiio.cdb[0],
13466			io->io_hdr.nexus.initid.id,
13467			io->io_hdr.nexus.targ_port,
13468			io->io_hdr.nexus.targ_target.id,
13469			io->io_hdr.nexus.targ_lun,
13470			(io->io_hdr.io_type ==
13471			CTL_IO_TASK) ?
13472			io->taskio.tag_num :
13473			io->scsiio.tag_num,
13474		        io->io_hdr.flags,
13475			io->io_hdr.status);
13476	} else
13477		io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13478#endif
13479
13480	/*
13481	 * This is an internal copy of an I/O, and should not go through
13482	 * the normal done processing logic.
13483	 */
13484	if (io->io_hdr.flags & CTL_FLAG_INT_COPY) {
13485		if (have_lock == 0)
13486			mtx_unlock(&ctl_softc->ctl_lock);
13487		return;
13488	}
13489
13490	/*
13491	 * We need to send a msg to the serializing shelf to finish the IO
13492	 * as well.  We don't send a finish message to the other shelf if
13493	 * this is a task management command.  Task management commands
13494	 * aren't serialized in the OOA queue, but rather just executed on
13495	 * both shelf controllers for commands that originated on that
13496	 * controller.
13497	 */
13498	if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)
13499	 && (io->io_hdr.io_type != CTL_IO_TASK)) {
13500		union ctl_ha_msg msg_io;
13501
13502		msg_io.hdr.msg_type = CTL_MSG_FINISH_IO;
13503		msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc;
13504		if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io,
13505		    sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) {
13506		}
13507		/* continue on to finish IO */
13508	}
13509#ifdef CTL_IO_DELAY
13510	if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13511		struct ctl_lun *lun;
13512
13513		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13514
13515		io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13516	} else {
13517		struct ctl_lun *lun;
13518
13519		lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13520
13521		if ((lun != NULL)
13522		 && (lun->delay_info.done_delay > 0)) {
13523			struct callout *callout;
13524
13525			callout = (struct callout *)&io->io_hdr.timer_bytes;
13526			callout_init(callout, /*mpsafe*/ 1);
13527			io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13528			callout_reset(callout,
13529				      lun->delay_info.done_delay * hz,
13530				      ctl_done_timer_wakeup, io);
13531			if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13532				lun->delay_info.done_delay = 0;
13533			if (have_lock == 0)
13534				mtx_unlock(&ctl_softc->ctl_lock);
13535			return;
13536		}
13537	}
13538#endif /* CTL_IO_DELAY */
13539
13540	STAILQ_INSERT_TAIL(&ctl_softc->done_queue, &io->io_hdr, links);
13541
13542#ifdef CTL_DONE_THREAD
13543	if (have_lock == 0)
13544		mtx_unlock(&ctl_softc->ctl_lock);
13545
13546	ctl_wakeup_thread();
13547#else /* CTL_DONE_THREAD */
13548	for (xio = (union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue);
13549	     xio != NULL;
13550	     xio =(union ctl_io *)STAILQ_FIRST(&ctl_softc->done_queue)) {
13551
13552		STAILQ_REMOVE_HEAD(&ctl_softc->done_queue, links);
13553
13554		ctl_process_done(xio, /*have_lock*/ 1);
13555	}
13556	if (have_lock == 0)
13557		mtx_unlock(&ctl_softc->ctl_lock);
13558#endif /* CTL_DONE_THREAD */
13559}
13560
13561void
13562ctl_done(union ctl_io *io)
13563{
13564	ctl_done_lock(io, /*have_lock*/ 0);
13565}
13566
13567int
13568ctl_isc(struct ctl_scsiio *ctsio)
13569{
13570	struct ctl_lun *lun;
13571	int retval;
13572
13573	lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13574
13575	CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0]));
13576
13577	CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n"));
13578
13579	retval = lun->backend->data_submit((union ctl_io *)ctsio);
13580
13581	return (retval);
13582}
13583
13584
13585static void
13586ctl_work_thread(void *arg)
13587{
13588	struct ctl_softc *softc;
13589	union ctl_io *io;
13590	struct ctl_be_lun *be_lun;
13591	int retval;
13592
13593	CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13594
13595	softc = (struct ctl_softc *)arg;
13596	if (softc == NULL)
13597		return;
13598
13599	mtx_lock(&softc->ctl_lock);
13600	for (;;) {
13601		retval = 0;
13602
13603		/*
13604		 * We handle the queues in this order:
13605		 * - task management
13606		 * - ISC
13607		 * - done queue (to free up resources, unblock other commands)
13608		 * - RtR queue
13609		 * - incoming queue
13610		 *
13611		 * If those queues are empty, we break out of the loop and
13612		 * go to sleep.
13613		 */
13614		io = (union ctl_io *)STAILQ_FIRST(&softc->task_queue);
13615		if (io != NULL) {
13616			ctl_run_task_queue(softc);
13617			continue;
13618		}
13619		io = (union ctl_io *)STAILQ_FIRST(&softc->isc_queue);
13620		if (io != NULL) {
13621			STAILQ_REMOVE_HEAD(&softc->isc_queue, links);
13622			ctl_handle_isc(io);
13623			continue;
13624		}
13625		io = (union ctl_io *)STAILQ_FIRST(&softc->done_queue);
13626		if (io != NULL) {
13627			STAILQ_REMOVE_HEAD(&softc->done_queue, links);
13628			/* clear any blocked commands, call fe_done */
13629			mtx_unlock(&softc->ctl_lock);
13630			/*
13631			 * XXX KDM
13632			 * Call this without a lock for now.  This will
13633			 * depend on whether there is any way the FETD can
13634			 * sleep or deadlock if called with the CTL lock
13635			 * held.
13636			 */
13637			retval = ctl_process_done(io, /*have_lock*/ 0);
13638			mtx_lock(&softc->ctl_lock);
13639			continue;
13640		}
13641		if (!ctl_pause_rtr) {
13642			io = (union ctl_io *)STAILQ_FIRST(&softc->rtr_queue);
13643			if (io != NULL) {
13644				STAILQ_REMOVE_HEAD(&softc->rtr_queue, links);
13645				mtx_unlock(&softc->ctl_lock);
13646				retval = ctl_scsiio(&io->scsiio);
13647				if (retval != CTL_RETVAL_COMPLETE)
13648					CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13649				mtx_lock(&softc->ctl_lock);
13650				continue;
13651			}
13652		}
13653		io = (union ctl_io *)STAILQ_FIRST(&softc->incoming_queue);
13654		if (io != NULL) {
13655			STAILQ_REMOVE_HEAD(&softc->incoming_queue, links);
13656			mtx_unlock(&softc->ctl_lock);
13657			ctl_scsiio_precheck(softc, &io->scsiio);
13658			mtx_lock(&softc->ctl_lock);
13659			continue;
13660		}
13661		/*
13662		 * We might want to move this to a separate thread, so that
13663		 * configuration requests (in this case LUN creations)
13664		 * won't impact the I/O path.
13665		 */
13666		be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13667		if (be_lun != NULL) {
13668			STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13669			mtx_unlock(&softc->ctl_lock);
13670			ctl_create_lun(be_lun);
13671			mtx_lock(&softc->ctl_lock);
13672			continue;
13673		}
13674
13675		/* XXX KDM use the PDROP flag?? */
13676		/* Sleep until we have something to do. */
13677		mtx_sleep(softc, &softc->ctl_lock, PRIBIO, "-", 0);
13678
13679		/* Back to the top of the loop to see what woke us up. */
13680		continue;
13681	}
13682}
13683
13684void
13685ctl_wakeup_thread()
13686{
13687	struct ctl_softc *softc;
13688
13689	softc = control_softc;
13690
13691	wakeup_one(softc);
13692}
13693
13694/* Initialization and failover */
13695
13696void
13697ctl_init_isc_msg(void)
13698{
13699	printf("CTL: Still calling this thing\n");
13700}
13701
13702/*
13703 * Init component
13704 * 	Initializes component into configuration defined by bootMode
13705 *	(see hasc-sv.c)
13706 *  	returns hasc_Status:
13707 * 		OK
13708 *		ERROR - fatal error
13709 */
13710static ctl_ha_comp_status
13711ctl_isc_init(struct ctl_ha_component *c)
13712{
13713	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13714
13715	c->status = ret;
13716	return ret;
13717}
13718
13719/* Start component
13720 * 	Starts component in state requested. If component starts successfully,
13721 *	it must set its own state to the requestrd state
13722 *	When requested state is HASC_STATE_HA, the component may refine it
13723 * 	by adding _SLAVE or _MASTER flags.
13724 *	Currently allowed state transitions are:
13725 *	UNKNOWN->HA		- initial startup
13726 *	UNKNOWN->SINGLE - initial startup when no parter detected
13727 *	HA->SINGLE		- failover
13728 * returns ctl_ha_comp_status:
13729 * 		OK	- component successfully started in requested state
13730 *		FAILED  - could not start the requested state, failover may
13731 * 			  be possible
13732 *		ERROR	- fatal error detected, no future startup possible
13733 */
13734static ctl_ha_comp_status
13735ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state)
13736{
13737	ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK;
13738
13739	printf("%s: go\n", __func__);
13740
13741	// UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap)
13742	if (c->state == CTL_HA_STATE_UNKNOWN ) {
13743		ctl_is_single = 0;
13744		if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
13745		    != CTL_HA_STATUS_SUCCESS) {
13746			printf("ctl_isc_start: ctl_ha_msg_create failed.\n");
13747			ret = CTL_HA_COMP_STATUS_ERROR;
13748		}
13749	} else if (CTL_HA_STATE_IS_HA(c->state)
13750		&& CTL_HA_STATE_IS_SINGLE(state)){
13751		// HA->SINGLE transition
13752	        ctl_failover();
13753		ctl_is_single = 1;
13754	} else {
13755		printf("ctl_isc_start:Invalid state transition %X->%X\n",
13756		       c->state, state);
13757		ret = CTL_HA_COMP_STATUS_ERROR;
13758	}
13759	if (CTL_HA_STATE_IS_SINGLE(state))
13760		ctl_is_single = 1;
13761
13762	c->state = state;
13763	c->status = ret;
13764	return ret;
13765}
13766
13767/*
13768 * Quiesce component
13769 * The component must clear any error conditions (set status to OK) and
13770 * prepare itself to another Start call
13771 * returns ctl_ha_comp_status:
13772 * 	OK
13773 *	ERROR
13774 */
13775static ctl_ha_comp_status
13776ctl_isc_quiesce(struct ctl_ha_component *c)
13777{
13778	int ret = CTL_HA_COMP_STATUS_OK;
13779
13780	ctl_pause_rtr = 1;
13781	c->status = ret;
13782	return ret;
13783}
13784
13785struct ctl_ha_component ctl_ha_component_ctlisc =
13786{
13787	.name = "CTL ISC",
13788	.state = CTL_HA_STATE_UNKNOWN,
13789	.init = ctl_isc_init,
13790	.start = ctl_isc_start,
13791	.quiesce = ctl_isc_quiesce
13792};
13793
13794/*
13795 *  vim: ts=8
13796 */
13797